Across the tech sector and beyond, we are watching a relentless wave of job cuts masquerading as inevitable efficiency. Roles are being slashed by the thousands and entire departments are being hollowed out, all under the convenient banner of AI-powered restructuring.

The sheer volume of people shown the door is stark, brutal, and frankly, alarming.

At the beginning of my career, I had the privilege of learning under some truly rock-solid mentors. Among the many lessons one of them shared with me, one piece of advice has stood out above the rest, a principle I continue to follow to this day.

Don’t worry about how much tax you pay. Earn more.

It was a brilliant reframing of how to look at growth. It forces you to stop obsessing over what you’re losing at the margins and start focusing entirely on expanding the pie.

Last week alone, we were hit with another barrage of news revolving around layoffs fueled by AI automation.

Subscribe now

Fair enough, we have heard the drill one too many times:

The narrative dictates that companies simply don’t need large teams anymore. Code can be generated, workflows automated, and operations streamlined. This message is increasingly becoming obvious.

Yet, instead of utilizing this newly found operational leverage to innovate, the industry has just been applying one band-aid over another, shrinking their way to temporary profitability.

This simply doesn’t add up.

I am curious to know what happened to the promised explosion in productivity and the subsequent revenue increases.

If AI is truly doing the heavy lifting, where is the massive surge in new products, better services, and top-line growth?

How long are the industry stalwarts going to keep focusing obsessively on the bottom line? Trimming expenses and firing talent might make the next quarterly earnings report look palatable, but you cannot shrink your way to greatness. The pressing question remains: when will the focus finally shift back to the entries on the top?

When do we get back to building, expanding, and simply earning more?

How to Actually “Earn More” with AI

If companies want to stop playing defensive accounting and start building empires again, the playbook needs to change. Here is what an obsession with the top line looks like when powered by AI:

• Accelerating R&D and Product Pipelines

  Instead of using AI just to write boilerplate code to save money, companies should use it to prototype ten times faster. The real winners will use AI to validate bold ideas, explore adjacent markets, and ship entirely new product lines that were previously too resource-intensive to even consider.

• Supercharging the Workforce, Not Replacing It

  A smaller team might save on salaries, but an AI-augmented team can tackle impossible problems. When you automate the drudgery, the smart move isn’t to fire the employee. The smart move is to redirect their newly freed brainpower toward strategy, high-touch customer relationships, and creative problem-solving. That is where premium value is generated.

• Deploying Hyper-Personalization at Scale

  We finally have the computational power to treat every single customer like they are our only customer. AI can analyze behavior and anticipate needs in real-time, allowing businesses to offer bespoke services. Highly personalized experiences don’t just reduce churn; they drive higher conversions, absolute loyalty, and increased lifetime value.

• Creating Net-New Revenue Streams

  AI isn’t just a tool to optimize the current business model; it is the foundation for new ones. Whether it is monetizing proprietary data insights, launching AI-driven advisory features, or building intelligent, self-serve tools for your own user base, the technology opens revenue doors that were previously locked.

My mentor was right: You can spend your entire life meticulously trying to optimize your tax bill, or you can channel that same energy into doubling your income.

The industry stalwarts celebrating their slashed payrolls are just optimizing their tax bills.

But the next generation of titans? They are looking at AI and figuring out exactly how to earn more.

It is time to decide which game we want to play.

Share Technoclast Insights

That is roughly how long it takes for a kitchen, restaurant, or factory floor to grind to a halt once the gas cylinder runs out.

• No flame.

• No cooking.

• No production.

Across India alone, hundreds of millions of LPG cylinders move through supply chains every year. Yet shortages still appear suddenly in cities and towns, leaving households and businesses scrambling.

The immediate reaction is usually frustration.

The more interesting question is this:

Did the shortage really come without warning?

In most cases, the signals were already there.

Demand was building.
Ordering patterns were shifting.
Seasonal events were approaching.

The supply chain simply failed to read the signals in time.

This edition explores how demand driven forecasting changes that equation and why planning for the unexpected has become a core capability for modern supply chain professionals.

Subscribe now

1. Demand Driven Forecasting

Most traditional supply chains follow a simple planning cycle.

1. Look at last month’s sales

2. Estimate next month’s demand

3. Place orders accordingly

This approach is often called forecast push planning. It relies heavily on historical averages.

The problem is obvious.

Planning based only on the past assumes the future will look similar.

Real markets rarely behave that way.

Demand driven forecasting takes a very different approach.

Instead of relying primarily on historical averages, it uses live demand signals to adjust forecasts continuously.

These signals often include:

• Point of sale data

• Distributor orders

• Customer bookings

• Market events

• Weather forecasts

• Local economic activity

The objective is simple.

Detect demand shifts early enough to act before shortages occur.

A simple analogy

Imagine a kirana store owner watching customer behavior closely.

She notices two consistent patterns:

• When school exams begin, families cook more meals at home

• During local festivals, gas refill requests rise by around 40 percent

Instead of waiting for inventory to run out, she increases orders in advance.

That is demand driven thinking.

Not reacting to shortages.
Preventing them.

2. The Gas Cylinder

At first glance, LPG demand looks predictable.

People cook every day.
Households refill cylinders regularly.

In reality, demand is influenced by several complex forces.

Share

1. Seasonal demand spikes

Consumption increases during certain periods.

Examples include:

• Winter months when households heat water more frequently

• Major festivals such as Diwali, Eid, Christmas, and Pongal

• Peak catering seasons

Even though these spikes occur every year, supply chains still struggle to plan for them effectively.

2. Volatile commercial demand

Restaurants and catering businesses introduce sudden demand surges.

Typical triggers include:

• Wedding seasons

• Large sports tournaments

• Food festivals

• Religious gatherings

Local consumption can jump 30 to 50 percent within days.

3. Invisible refill clusters

Many housing complexes follow synchronized refill cycles.

For example:

• Residents order refills on the first day of each month

• Distributors experience a sudden surge in demand

Without visibility into these patterns, shortages appear quickly.

4. Rapid growth in new users

Urban expansion introduces thousands of new LPG connections in short periods.

When supply chains designed for steady demand face sudden growth, inventory buffers disappear rapidly.

3. When the Unexpected Arrives

The concept of the black swan was popularized by Nassim Nicholas Taleb.

A black swan event has three characteristics:

1. It falls outside normal expectations

2. It arrives with little warning

3. It creates disproportionate impact

In supply chains, black swans expose hidden fragility.

Common black swan disruptions in gas supply chains

1. Industrial accidents

   • A fire or safety incident shuts down a refilling plant

   • Entire districts lose supply capacity

2. Extreme weather events

   • Floods or cyclones disrupt transport routes

   • Demand rises exactly when logistics fail

3. Fuel substitution shocks

   • Electricity price spikes push households back to LPG

4. Pandemic driven consumption shifts

   • During lockdowns, home cooking increases dramatically

5. Transport disruptions

   • Truck strikes or infrastructure failures block deliveries

The important insight is this:

Individual disruptions are unpredictable.

Disruption itself is predictable.

4. How Does It Help?

Demand driven forecasting is not just a planning tool.

It acts as an early warning system.

1. Real time demand sensing

Continuous monitoring detects anomalies quickly.

For example:

• A sudden 60 percent increase in cylinder orders in a region

• Demand rising above seasonal norms

This allows planners to respond days or weeks earlier.

2. Intelligent buffer stock design

Not all markets behave the same.

Demand driven models calculate buffer levels based on:

• Demand volatility

• Customer density

• Supply lead times

Urban distributors often require larger buffers than rural ones.

3. Multi source supplier networks

Forecasting models reveal supply bottlenecks.

Organizations can then build redundancy through:

• Multiple refilling plants

• Alternative transport routes

• Regional distribution hubs

Redundancy is not inefficiency.

It is insurance.

4. Scenario simulation and stress testing

Advanced planning systems allow teams to test scenarios such as:

• Demand doubling within 30 days

• Supplier disruptions for two weeks

• Transport cost increases of 25 percent

Running these simulations before a crisis ensures teams already know how to respond.

5. Collaborative demand visibility

A gas cylinder travels through several supply chain nodes:

1. Refinery

2. Bottling plant

3. Regional depot

4. Distributor

5. Customer

Demand driven forecasting connects these nodes through shared data signals.

When distributor orders surge, upstream suppliers see it immediately.

This synchronization dramatically improves responsiveness.

5. Practical Steps

Restaurants, bakeries, hospitals, and catering businesses cannot afford sudden gas shortages.

Several simple practices can reduce operational risk.

1. Track consumption patterns

Maintain a weekly usage log.

Within a few months, patterns usually become visible.

2. Maintain a safety buffer

Never operate with zero reserve cylinders.

A simple rule:

• Reorder when the last cylinder is opened, not when it is empty.

3. Diversify suppliers

Relying on a single distributor creates unnecessary risk.

Maintaining relationships with two suppliers provides backup.

4. Communicate demand early

Inform distributors in advance about:

• Large catering orders

• Seasonal spikes

• Special events

Demand information often travels too slowly in supply chains.

Early communication helps everyone plan better.

5. Participate in forecasting programs

Some distributors offer collaborative demand forecasting programs.

Businesses that share forward demand estimates often receive supply priority during shortages.

6. Antifragile Supply Chains

There is a concept beyond resilience.

Antifragility.

An antifragile system does not just survive shocks.

It becomes stronger because of them.

In an antifragile gas supply chain:

• Shortage events feed new data into forecasting models

• Buffer policies improve continuously

• Alternative suppliers are qualified

• Transport redundancy expands

• Demand sensing becomes sharper

Each disruption becomes a learning event.

Industries such as pharmaceuticals, food manufacturing, and energy distribution are already building supply chains this way.

The tools exist.

The data exists.

What is often missing is the willingness to invest in planning before a crisis occurs.

Key Takeaways

1. Demand driven forecasting replaces historical planning with real time demand signals.

2. Gas cylinder supply chains appear simple but hide complex demand dynamics including seasonal spikes and commercial surges.

3. Black swan disruptions cannot be predicted individually, but the presence of disruption is inevitable.

4. Combining demand sensing, buffer stock strategies, multi source suppliers, and collaborative forecasting creates resilient supply chains.

5. Businesses can start small by tracking consumption, maintaining buffers, diversifying suppliers, and sharing demand information early.

Closing Thought

A gas cylinder shortage may appear like a minor inconvenience.

In reality, it reveals a much larger supply chain truth.

When planning assumes tomorrow will look like yesterday, systems become fragile.

Demand driven forecasting does not claim to predict every disruption.

What it does provide is the ability to:

• Detect change earlier

• Respond faster

• Build supply chains that do not collapse when the unexpected arrives

Because the real question is never whether disruption will happen.

Only when.

Artificial Intelligence is a massive financial bubble ready to burst. But this anxiety fundamentally misunderstands the difference between a soaring stock market and a relentless technological march.

Think back to the great railroad boom of the 19th century.

Investors poured unprecedented fortunes into railway companies, driving stock prices to dizzying and unsustainable heights. When reality eventually set in, the financial frenzy collapsed spectacularly, wiping out fortunes and bankrupting overleveraged companies overnight.

But here is the crucial part - The tracks were already laid. The steam engines did not stop running. Those very rails went on to permanently transform transportation, commerce, and human society. The crashing stock prices did not change the physics of the locomotive.

Today, we are watching a similar story unfold with AI. It is vital to separate the economic speculation from the underlying technological reality.

Subscribe now

The Economic Phenomenon

A financial bubble is a market condition where the price of an asset rises significantly above its fundamental value. In the AI sector, we are certainly seeing signs of this exuberance.

• Valuation Extremes

  Billions of dollars are flowing into startups and tech giants alike, often detached from current revenues or immediate profitability.

• Capital Concentration

  A staggering percentage of global venture capital is currently directed solely toward AI ventures, creating a highly skewed investment landscape.

• The Fear of Missing Out

  Investors are pouring money into the space not necessarily because they understand the underlying models, but because they are terrified of being left behind.

This financial froth means that a market correction is highly likely. Many AI startups will fail. Valuations will normalize. But a correction in the market is not a cancellation of the future.

The Steadfast Technological Advancement

While the stock tickers fluctuate wildly, the actual science of artificial intelligence is advancing at a steady and measurable pace.

Unlike pure financial speculation, the current AI boom is backed by massive, tangible infrastructure.

• Physical Infrastructure

  The world’s largest companies are building real, capital intensive data centers and deploying millions of advanced semiconductors. These are concrete assets, not just speculative ideas.

  +1

• Accelerating Capability

  The underlying models continue to become more efficient, multimodal, and capable of complex reasoning. Code generation, medical research, and logistics are already seeing measurable productivity gains.

• Broad Adoption

  Businesses across the globe are integrating these tools into their daily workflows, shifting AI from an experimental novelty to a core operational requirement.

So, essentially

A financial bubble popping clears out unsustainable business models and speculative froth. It does not erase the technological breakthroughs that caused the excitement in the first place. When the dot com bubble burst in the early 2000s, many web companies vanished, but the internet itself continued to grow and eventually redefined the global economy.

AI is walking the exact same path. The economics will inevitably experience turbulence, but the steadfast advancement of the tech is already locked in. The tracks are being laid, and the train is moving forward.

Share Technoclast Insights

$270 billion in cross-border investments in a single year. 219 Gigawatts of projected power demand.

For centuries, the maps of global power were drawn with ink and blood over natural resources, warm-water ports, and arable land. Today, those maps are being redrawn over windowless, climate-controlled warehouses. Data centers are no longer just the passive plumbing of the internet. They are the most vital strategic strongholds of the 21st century.

If we want to understand the future of global security, we need to fundamentally rethink what constitutes a military base.

Here is why the global power struggle has quietly relocated to the cloud.

Subscribe now

Compute as the New Deterrent

In the 20th century, nuclear stockpiles acted as the ultimate geopolitical deterrent. Today, that deterrent is shifting toward Sovereign AI, which is artificial intelligence developed, trained, and hosted entirely within a single jurisdiction.

You cannot run predictive battlefield analytics, defend against automated cyber warfare, or develop autonomous defensive systems without hyperscale computing. The capacity to process exabytes of data is no longer a commercial advantage; it is a matter of national survival. A nation without its own compute infrastructure is entirely reliant on the goodwill of those who host its data, rendering it strategically defenseless

Choke Points & Invisible Blockades

Historical blockades involved warships cutting off trade routes. Today’s blockades are economic and technological.

A modern data center relies on highly advanced semiconductors, specialized cooling hardware, and the sprawling network of underwater fiber-optic cables that connect them to the world.

When export controls are placed on advanced microchips, it is not merely a trade dispute. It is a modern siege tactic designed to starve a rival’s infrastructure of processing power. Control over the supply chain of a silicon wafer is now just as critical as control over a major physical shipping strait.

The Weaponization of Watts

The scale of energy required by these facilities is staggering. By 2030, global data centers are projected to consume more electricity than many developed nations.

This creates a terrifying new vulnerability. To maintain technological sovereignty, a state must secure uninterrupted, massive supplies of energy and water for cooling. Consequently, an adversary no longer needs to target a military installation to cripple a nation. Disrupting the regional power grid or water supply that feeds a primary data center cluster can blind intelligence networks and freeze financial systems in seconds. Energy policy is now inextricably linked to digital defense.

Share

The Vulnerability of Concentration

We are currently building massive, centralized points of failure. In any future kinetic or cyber conflict, severing a region’s connection to the cloud will be a primary, day one objective.

Why risk human lives and expensive traditional weaponry when a coordinated cyberattack, or a targeted strike on a massive server farm, can simultaneously disable an adversary’s logistics, banking, and communications? The concentration of digital assets into massive, centralized warehouses makes them high-value targets that are increasingly difficult to defend.

A Paradigm Shift in Power

We need to start educating our policymakers, strategists, and citizens about this new reality. Geopolitical dominance is no longer measured purely by the number of aircraft carriers in a fleet. It is measured by

• megawatts, compute capacity, and network latency.

The physical infrastructure of the internet is the new contested territory. The nations and alliances that build, secure, and power these facilities will dictate the technological and economic realities for the rest of the world.

The cold war of our era is already here, and the servers are running hot.

Share Technoclast Insights

As we navigate the complex intersections of Agentic AI, Data Strategy, and Governance, it’s becoming clear that the threats we face are scaling at the exact same speed as our innovations.

In the world of financial forensics and revenue recovery, particularly when tracking complex fraud across the diverse markets of the Asia Pacific region, the most dangerous threats rarely look like a traditional bank heist.

Instead, they look like millions of microscopic anomalies, perfectly camouflaged within massive streams of legitimate, automated transactions.

Subscribe now

We are now witnessing this exact same dynamic in how software is built. This week, the tech industry received a massive wake-up call in the form of hackerbot-claw.

This wasn’t a human hacker typing furiously in a dark room; it was an autonomous AI agent running 24/7. Over a seven-day campaign, it systematically scanned, verified, and compromised the digital assembly lines of major open-source projects, including those from Microsoft, DataDog, and the Cloud Native Computing Foundation (CNCF).

Whether you write code for a living, manage the teams that do, or lead the business strategy that relies on that software, understanding this automated assault is mandatory. Here is a look at how this happened, minus the dense technical jargon, and the governance required to stop it

Hacking the Process

When we think of cyberattacks, we usually picture someone breaking into a live application to steal customer data. Hackerbot-claw did something different: it attacked the factory where the software is made.

In modern development, teams use automated pipelines (often called CI/CD) to test, build, and deploy code. These pipelines are highly privileged. They hold the keys to the kingdom so they can do their jobs efficiently. The bot realized it didn’t need to break into the final product if it could just trick the factory robots into handing over the keys.

It used a few brilliantly creative, yet fundamentally simple, tactics:

• The Trojan Horse

  The bot submitted seemingly harmless code updates. However, it hid malicious instructions inside the routine automated quality checks. When the system automatically picked up the code to test it, it accidentally executed the attacker’s commands, inadvertently handing over high-level access tokens.

• The Poisoned Label

  Imagine a factory scanner that reads barcodes. Now imagine if the barcode itself contained a command that reprogrammed the scanner. The bot disguised malicious commands within file names and branch names. When the automated system tried to read the name, it was tricked into running the hidden code instead.

• The AI Con Artist

  In the most novel attack, the bot targeted an AI code-reviewer acting as a gatekeeper for one of the projects. The attacker altered the underlying rulebook that the AI relied on, attempting to socially engineer the AI into approving malicious changes and covering its tracks. (Fortunately, the underlying AI model was smart enough to refuse the manipulation).

  Share

Redefining Governance

Defending against a tireless, automated bot requires a strategic mindset. It demands a level of corporate stoicism, acknowledging the chaos of relentless, AI-driven threats without panic, and calmly fortifying our digital perimeters. We cannot rely on manual, human intervention to fight machine-speed attacks.

Here are the critical governance guardrails every organization must adopt:

1. Enforce Zero-Trust Automation

We can no longer give automated systems blind, sweeping access. Just as you wouldn’t give a single employee the keys to every vault in a bank, an automated testing process should only have the exact permissions it needs to perform its specific task, and nothing more. If a system only needs to read data, explicitly block its ability to write or alter it.

2. Audit the Supply Chain Inputs

Every piece of data that enters a development pipeline, whether it’s a line of code, a file name, or an automated command must be treated as untrusted until verified. We must implement rigorous sanitization checks to ensure that labels aren’t secretly carrying executable commands.

3. Quarantine Untrusted Activity

Automated processes should never be allowed to automatically execute code submitted by unverified or external sources without a human-in-the-loop or a highly restricted, isolated testing environment.

4. Govern the AI Context

As we integrate more Agentic AI into our workflows, the instructions and context we feed these models become critical infrastructure. The rulebooks that guide our internal AI agents must be locked down and monitored just as heavily as our most sensitive databases.

We are building the next generation of software with intelligent agents, but we must ensure we aren’t leaving the back door open for their malicious counterparts. The future belongs to those who innovate rapidly, but govern wisely.

Until next time, stay secure.

Share Technoclast Insights

Back in the early days of machine translation, engineers famously fed a system the phrase out of sight, out of mind. The system translated it into Russian &then back to English.

Subscribe now

While AI has improved exponentially since those early days, the underlying lesson remains identical. Translating the words is not the same as understanding the entire ecosystem.

When Anthropic announced that its Claude Code AI could seamlessly read and translate 67-year-old COBOL into modern languages, the market reacted as if the entire mainframe industry had been rendered obsolete overnight. Investors hit the panic button, triggering a massive selloff of IBM stock.

However, as industry veterans quickly pointed out, translating a legacy language is only a fraction of the actual modernization battle.

Anatomy of an Overreaction

It is completely understandable why the market panicked initially. COBOL is the invisible backbone of the global economy, quietly powering 95 percent of ATM transactions and 80 percent of in-person credit card swipes. Because the original developers are retiring and universities no longer teach the language, maintenance costs are skyrocketing.

An AI tool that promises to automate this massive technical debt sounds like a fatal blow to the consulting revenues of legacy tech giants. However, analysts at Reuters, VentureBeat, and The Futurum Group rapidly poured cold water on the hype.

The fundamental error driving the market panic is the conflation of code translation with platform modernization. Anthropic Claude Code can absolutely map dependencies and translate legacy COBOL syntax into Java or Python. But IBM has already been offering a similar AI tool, watsonx Code Assistant, since 2023. The true bottleneck in mainframe modernization has never been just rewriting the code.

The Real Bottlenecks of Modernization

Translating syntax is fast, but deploying an enterprise-grade financial system is slow, methodical, and heavily reliant on human expertise. Here is why the migration process remains human-led.

• Hardware Coupling

  Decades-old COBOL applications are tightly integrated with the underlying mainframe hardware to achieve high-speed transaction processing. Simple code translation cannot replicate this optimized performance.

• Behavioral Equivalence

  Financial institutions require rigorous validation. A modernized system must process billions of transactions exactly the same way the old system did, down to the microsecond, to avoid catastrophic accounting errors.

• Regulatory Audits

• AI cannot sit in a room with government regulators to explain compliance protocols. Every step of a financial migration requires extensive audit trails and human accountability.

• Ecosystem Integration

  Legacy code does not operate in a vacuum. It interacts with thousands of external databases and partner systems that also need careful, strategic realignment.

  Share

Translation vs. Modernization

Your Next Strategy Meeting

Here are a few core takeaways to keep your team grounded when the next AI disruption narrative hits the news cycle.

• AI can change the syntax, but human engineers must manage the system.

• A translated codebase without structural integration is simply a fast track to a broken ecosystem.

• The stock market loves a disruption narrative, but enterprise reality runs on audits, security, and compliance.

In Essence

The recent Anthropic announcement is a phenomenal showcase of how rapidly AI coding capabilities are advancing.

Tools that automate exploration and analysis will undeniably save thousands of hours in legacy migrations. But until an AI can sign off on a compliance audit or seamlessly untangle a 60-year-old hardware dependency, true modernization will remain a human-directed endeavor.

Do not let a sell first, ask questions later market trend distract you from the complex fundamentals of enterprise engineering.

Share Technoclast Insights

This view misses a crucial point.

Modern AI, especially agentic AI, behaves in ways that cannot be explained by examining its code line by line.

In this edition, let me share why AI can act in complex, emergent, and strategic ways.

The Misconception

Thinking AI is a Giant If-And-Else Machine.

For a long time, AI was simple. Early programs followed rules like this:

• If the user asks X, do Y.

• Otherwise, do Z.

Modern AI works very differently. It is trained on massive amounts of data and learns patterns, rather than following explicit instructions for every possible situation. For example, a language model like GPT-5 mini was not programmed with responses for every question. Instead, it was trained on billions of examples to learn how language works. The instructions written by developers define how the AI learns, not the exact words it will produce.

Emergence

Behaviour beyond Programming

AI can display behavior that was never explicitly programmed. Examples include:

• Summarizing text and generating analogies.

• Negotiating or strategizing to complete multi-step tasks.

• Planning several steps ahead to reach a goal without being told exactly how.

This behavior arises from learning patterns in data and using optimization algorithms to achieve objectives.

Humans offer a useful analogy. Our DNA does not contain instructions to invent calculus or play the violin, yet humans can do these things. AI can similarly develop skills that were not directly programmed.

What is Agentic AI?

Agentic AI refers to systems that can:

• Set sub-goals

• Plan multiple steps ahead

• Adjust behavior based on feedback from their environment

These systems are not just following rules. They act like agents that make decisions and pursue objectives.

Technical Perspective

Agentic AI often combines:

1. Large Pretrained Models trained on extensive datasets to learn patterns.

2. Goal-Oriented Reinforcement Learning, where the AI is rewarded for achieving outcomes rather than following fixed rules.

3. Planning and Memory Modules that allow reasoning across multiple steps and predicting consequences before acting.

For instance, a warehouse robot with agentic AI does more than follow a path. It observes obstacles, predicts worker movements, and plans the most efficient route to pick up packages without human guidance.

Why AI is Not Just Following Instructions

Even though AI is designed with certain goals, how it achieves them is not predetermined. Key points include:

1. Complexity is Non-Linear
   Small changes in input can produce very different outputs.

2. Learning Creates Novel Patterns

   AI can generate solutions that were not in its training data.

3. Multi-Step Reasoning

   Agentic AI can chain decisions, producing behavior that may seem creative or strategic.

A chess engine illustrates this. The rules of chess are fixed, but the strategies it develops are discovered, not programmed.

An Example

A simple analogy is teaching a child by showing them thousands of examples of stories, conversations, and problem-solving.

You do not write a script for what they will say. Over time, they learn patterns and begin generating original responses. Modern AI works in a similar way. It is constrained by goals and data, but its behavior is not fully determined by explicit instructions.

Why This Matters?

Understanding this distinction is important because it:

• Explains why AI can be unpredictable

• Shows why agentic AI requires careful oversight, especially when interacting with humans or real-world systems

• Helps anticipate both opportunities and risks, instead of underestimating AI by assuming it is a simple rule-follower

Closing Thought

Saying AI just follows instructions is like saying a jazz musician just follows sheet music. The framework exists, but the improvisation and creativity that emerge go far beyond simple programming. Agentic AI can learn, adapt, and act in ways that surprise even its creators.

AI is not magic and not sentient yet, but it is also far more than a set of rules. It is a complex system capable of acting, adapting, and generating outcomes that cannot be traced line by line to a programmer’s code.

The fraud only surfaced when a government department tried to close its account, and the numbers simply didn’t add up.

By then, it was too late.

• Suspensions

• Forensic audits

• Regulatory scrutiny

• A stock in freefall

The institution’s own leadership acknowledged it plainly:

The entire maker-checker-authorizer system was in place. And yet, a group of people came together and made the fraud happen anyway.

That admission should stop every risk professional, compliance officer, and banking executive cold.

Not A Technology Failure.

It is a control failure engineered by humans who understood the controls a tad too much.

Collusive fraud is the most dangerous species of financial crime precisely because it defeats conventional safeguards from the inside. When the maker, checker, and authorizer are all compromised, or when even one trusted insider manipulates the chain, rule-based systems, periodic audits, and manual reconciliation become theatre. The fraudsters don’t break the system. They operate it.

This is not unique to one institution or one branch. Insider threat and internal collusion are consistently among the top drivers of fraud losses globally, across banking, insurance, supply chain, and public finance. The Association of Certified Fraud Examiners estimates that organizations lose 5% of annual revenues to fraud, with a significant share involving internal actors.

The uncomfortable truth: most of these frauds leave data trails. They just aren’t being read in real time.

Subscribe now

Present or Future?

Data and AI are not the future of fraud risk management. They are the present, and most institutions are still catching up.

Here is what modern AI-powered fraud and risk systems can do that legacy controls simply cannot, and how they must be deployed with intent, not just installed with hope.

1. Behavioral Anomaly Detection: Move Beyond Rules, Into Patterns

Rule engines are reactive. They catch what you already know to look for. AI learns what normal looks like across every user, account, branch, and transaction relationship, and raises an alert the moment something diverges, even if no rule has been broken. An employee processing large-value instructions at unusual hours. A branch where reconciliation exceptions are always cleared by the same two people. A cluster of approvals that moves just below every threshold. These signals are invisible to policy manuals. They are loud to a well-trained model.

The imperative: deploy behavioral AI not as a monitoring tool sitting in a back office, but as a live risk layer embedded into every approval workflow.

2. Graph Intelligence to Surface Collusion: See the Network, Not Just the Transaction

Collusion is, at its core, a network problem. And networks have signatures. Graph-based AI can map every relationship between employees, accounts, beneficiaries, approvers, and external entities, and detect when those relationships form patterns associated with coordinated fraud: shared access clusters, unusual co-authorization chains, money flows that loop back through the same nodes. What a human auditor sampling 5% of transactions cannot see, a graph neural network analyzing 100% of the transaction universe sees in real time.

The imperative: invest in graph-based analytics infrastructure and connect it to your identity, access, and payments data. Fraud hides in relationships. So must your detection capability.

3. Continuous Surveillance: End the Audit Window, Permanently

The gap between audits is where fraud lives. Monthly, quarterly, and annual reviews are retrospective by design, and fraudsters with institutional knowledge know exactly how to operate within that window. AI enables continuous, always-on monitoring of every transaction, every access log, every exception, every override, with risk scores updated in real time and anomalies escalated immediately.

The imperative: retire the mindset that surveillance is something you do periodically. Build infrastructure where the system watches always, so your people can investigate always.

4. Unsupervised Learning for Fraud You Haven’t Seen Before

The fraud method in this case, forged physical cheques, was acknowledged by the institution’s own leadership as the oldest fraud in banking. Yet it went undetected for a prolonged period. Supervised models only catch what they have been trained on. Unsupervised models surface statistical outliers regardless of fraud typology, because they don’t need to know what the fraud is, only that something is statistically out of place.

The imperative: complement your supervised fraud models with unsupervised anomaly detection running in parallel. Unknown fraud patterns are only unknown until someone looks at the right signal.

Tech Is Just The Beginning

Technology alone is never enough. And this is the part of the conversation the industry consistently avoids.

Deploying AI in fraud risk is not like installing software. It demands a profound shift in how your people think, what they know, and how they engage with data every single day.

Risk and Compliance Teams Must Become Data-Literate, Not Optionally, But Urgently

The era of the fraud investigator who works purely from instinct, experience, and sampled transaction reports is ending. Today’s risk professional must be able to read a model output, interrogate an anomaly score, understand why an alert was raised, and make a judgment call that combines human context with machine signal. This is not a data science skill. It is a risk professional skill, and it needs to be built deliberately through structured learning, cross-functional exposure, and hands-on engagement with the tools.

Learn SQL. Understand what a confusion matrix tells you about your fraud model. Know the difference between a false positive rate and a precision score, and why both matter in your context. These are not optional anymore.

Data Scientists and AI Engineers Must Learn the Domain, Deeply

A model built without genuine understanding of how banking operations work, how fraud typologies evolve, how internal controls are structured, and where the human vulnerabilities lie, will produce alerts that operations teams ignore within six months. The technical and domain communities must work in far closer proximity than most organizations currently allow. Embed your AI team in risk. Embed your risk team in AI development. Shadow each other. Build together. The boundary between these disciplines is where the best fraud detection capability is forged, and where most institutions are currently failing.

Leave a comment

Leaders Must Create the Conditions for AI to Actually Work

AI systems will surface uncomfortable things about trusted people, long-standing processes, and high-performing branches or teams. The instinct to explain away an alert about a senior or well-regarded employee is powerful and deeply human. Leadership must actively and visibly build a culture where every alert is treated with the same rigor regardless of who it involves, where the model’s signal is never suppressed for political convenience, and where acting on AI-generated intelligence is rewarded, not quietly discouraged. Without this, even the most sophisticated fraud AI becomes very expensive decoration.

Skilling Up Is Not a One-Time Training Event. It Is a Continuous Operating Discipline.

Fraud evolves. Models drift. New typologies emerge. The institutions that win this fight are the ones that invest in continuous learning: regular model retraining, adversarial red-teaming, ongoing upskilling of risk analysts, and a culture of intellectual curiosity about what the data is trying to say. Attend the courses. Get the certifications. Run internal war-games where your risk team tries to defeat your own fraud models. Build learning loops, not just systems. Treat your people’s AI fluency as a risk asset that requires the same maintenance and investment as the technology itself.

Lastly..

The best AI has to offer is not efficiency. It is the ability to see what human systems are designed, or incentivized, to miss.

But it only delivers that capability when the humans working alongside it are skilled enough, empowered enough, and genuinely curious enough to act on what it reveals.

The real questions every institution should be asking today are not whether they have an AI fraud solution. The real questions are: Do our risk teams know how to work with it? Do our data teams understand the fraud domain well enough to build it right? And do we have a culture courageous enough to act on what it finds, even when that is uncomfortable?

If the answer to any of these is uncertain, that is the gap to close, before the next case, not after.

• The tools exist.

• The data exists.

• The science exists.

What the moment demands is people skilled enough to wield these tools with precision, leaders bold enough to act on their output, and institutions brave enough to let the data speak, even when it says something no one wants to hear.

Insider fraud is not a new problem.

Our capacity to detect and prevent it has never been greater. The only question is whether we build the human capability to match it.

Share Technoclast Insights

But recently, revenue authorities uncovered something much harder to swallow, a staggering ₹70,000 crore tax evasion scam, served hot from a routine restaurant inspection.

The recipe for this fraud?

Over 1 lakh restaurants manipulated their billing software to quietly delete cash transactions before filing their returns. It took an intensive investigation, the deployment of AI, and the parsing of 60TB of data to finally catch the scent.

Headlines are painting this as a massive win for data analytics. But if we pull back the curtain on public sector revenue management, a more uncomfortable truth emerges.

Subscribe now

The Post-Facto Fallacy

When you uncover evasion of this magnitude post-facto, more than half the battle is already lost.

Identifying a ₹70,000 crore hole years after the money has vanished isn’t a victory; it’s a financial autopsy.

Recovering dispersed, hidden, or spent capital is a massive drain on government resources, often yielding pennies on the dollar.

In my work leading Financial Forensics & Intelligence initiatives across the Asia Pacific region, my core mandate is to empower government and state authorities to prevent exactly this type of systemic leakage.

By arming tax administrations with proactive, forensic-grade Data & AI insights, we shift the paradigm from reactive audits to proactive, surgical intelligence. The result?

Usually billions in actual revenue recovery, secured before the funds ever leave the ecosystem.

We cannot afford to keep building better nets just to catch the fallout. We need predictive infrastructure that stops the leak before the drop falls.

Proactive Management & Financial Forensics

When 1 lakh entities coordinate to hit delete on their ledgers, standard auditing fails. Here is how Data & AI must be engineered to transform tax administration from a reactive trap into a proactive, forensic engine:

1. Managing Taxpayer Data

The Bedrock of Intelligence Fraud thrives in data silos.

The biryani scam worked because point-of-sale data wasn’t talking to broader financial networks. A cohesive data strategy begins with unified ingestion. By integrating GST networks, digital payment gateways, property records, and vendor supply chains into a centralized data lakehouse, we create a single, immutable source of truth. AI steps in here to sanitize data, handle fuzzy matching across disparate ID systems, and stitch together a comprehensive 360-degree financial profile.

2. Establishing Behavioral Patterns & Tracking Deviances

Static rules are easily bypassed by anyone who understands where the tripwires are hidden. We must move toward adaptive AI oversight. By establishing hyper-personalized baselines for every taxpayer, machine learning models continuously monitor operational behavior. Instead of waiting for a quarterly batch-run, these systems flag anomalous patterns, like a sudden spike in cancelled invoices or a divergence between utility consumption and reported output in real-time.

3. AI-Driven Financial Forensics: Reconstructing the Shadow Ledger

Data doesn’t just vanish. Financial forensics is the art of untangling the web to reconstruct the “shadow ledger.” When businesses attempt to erase digital paper trails, advanced forensics utilizes Natural Language Processing (NLP) to parse unstructured data—emails, contracts, and audit logs—flagging discrepancies between internal communications and official filings. It reconstructs the timeline of financial events to prove intent, transforming a suspicion of deleted data into an evidentiary trail that holds up in a tribunal.

4. Unmasking Collusions & Tracing Beneficial Ownership

Evasion at the ₹70,000 crore scale is rarely a lone-wolf operation. It requires syndicates. AI-driven graph network analysis is the ultimate forensic tool here.

It maps complex relationship webs to uncover both external and internal collusions. Externally, it peers through layers of shell companies to identify the Ultimate Beneficial Owner (UBO), tracking illicit capital flights across borders. Internally, it can track anomalies within the tax administration itself flagging if specific officers consistently approve unusual deductions, or if access logs show inappropriate modifications to taxpayer records.

5. Prioritizing to Maximize Recovery Yield

Not every red flag deserves the same deployment of forensic resources. If an AI system flags 50,000 anomalies, how do you decide where to send your human auditors first?

Predictive models assess the propensity to pay and the likelihood of recovery. By evaluating the scale of the evasion against the complexity of the audit and the asset liquidity of the evader, AI ranks cases by their projected ROI. This ensures public sector resources are surgically deployed where they will yield the maximum actual recovered revenue, not just the highest theoretical assessment.

6. Intelligent Enforcement and Tailored Recovery

This is perhaps the most critical component of applied AI ethics in public governance. When we look at compliance, we must acknowledge that not all non-compliance is syndicated malice. Sometimes, defaults stem from systemic friction, complex regulatory burdens, or temporary operational struggles.

Instead of swinging a blunt enforcement hammer, like blanket account freezes at every flag, intelligent forensic systems trace the anomaly to understand the why.

For genuine struggles, the system carves out personalized recovery paths:

• Automated nudges

• Tailored payment installment plans, and

• Proactive compliance assistance.

We reserve the heavy forensic artillery, deep-dive audits and legal asset seizure for calculated, large-scale evasion. Needless to say, optimzation principles must be deeply baked into ensure where we put in our efforts to maximize the recovery.

Managing the Kitchen

Innovation in public sector revenue isn’t just about deploying a shiny new algorithm to audit old data.

It’s about systemic integration, continuous behavioral monitoring, and deep financial forensics that protect the public purse ethically and effectively.

We need to stop trying to un-cook the biryani, and start managing the kitchen better. If not for anything, for the beloved Biryani’s sake!

Link to the News Article: https://www.hindustantimes.com/india-news/how-a-biryani-joint-check-in-hyderabad-uncovered-a-rs-70-000-crore-tax-evasion-scam-across-india-101771478527422.html

Share Technoclast Insights

And yet, paradoxically, the biggest danger today isn’t AI itself.

It’s overconfidence in its capabilities, a confidence that often forms without the operational grounding needed to deliver real value.

Subscribe now

Overconfidence is the Hidden Threat

Organizations are increasingly building strategies and making investments based on beliefs like:

• If we adopt the latest Large Language Model (LLM), our processes will magically improve.

• AI will totally replace human decision-making.

• We need only a minimal pilot before scaling everywhere.

This isn’t optimism. It is overconfidence bias, the psychological tendency to overestimate one’s own knowledge, ability, or level of control.

When paired with powerful AI narratives and boardroom pressure, this cognitive bias becomes organizational policy, resulting in the following

• Unrealistic expectations

• Budget overruns

• Misaligned deliverables

• Poor decisions based on automated outputs

The result? Many companies find themselves investing millions only to realize their problems weren’t solved or, worse, were exacerbated.

Overconfidence - What Are The Root Causes?

Overconfidence in AI doesn’t occur in a vacuum. It evolves at the intersection of technology hype, human psychology, and organizational incentives.

1. The Hype Cycle Takes Over

Tech press, analysts, and early adopters often highlight best-case success stories.

AI diagnosing medical images, autonomous systems, generative code, 80% productivity uplift claims. That creates a perception that AI is magically effective everywhere.

Without equally visible discussions on limitations, assumptions, failed pilots, and required expertise, decision-makers fall into the trap that AI will solve our entire problem set.

2. Lack of AI Literacy

GPT-like models generate plausible responses even when wrong. Business leaders lacking technical context tend to

• Overinterpret AI outputs

• Believe automated results are true

• Ignore error rates or uncertainty metrics

This illusion of certainty breeds organizational confidence without technical grounding.

3. Technical Debt and Process Gaps Mask Reality

AI’s effectiveness depends on

• Quality of data inputs

• Integration with existing systems

• Clear performance metrics

• Strong maintenance processes

What organizations often have instead is:

• Fragmented data

• Siloed IT

• Manual interventions

• No monitoring or governance

That gap between expectations & reality compounds as time goes on.

Subscribe now

4. Resistance Gets Overshadowed

This is the human side of transformation:

• Teams are afraid to challenge AI outputs

• Incentives tied to AI adoption rather than measured outcomes

• Leaders misreading enthusiasm for effectiveness

Overconfidence here becomes culture reinforcement, not thoughtful evaluation.

And, the Consequences

1. Escalating Tech Debt

AI systems can hide linearly growing costs:

• Off-the-shelf models require custom tuning

• APIs incur per-use costs

• Drift and retraining requirements rise

Without planning, what looked like a cheap solution becomes expensive technical overhead.

2. Poor Business Decisions

AI hallucinations and incorrect predictions don’t announce themselves.

Leaders without a data science grounding can:

• Treat outputs as facts

• Ignore measures of uncertainty

• Amplify errors into strategic choices

This leads to lost revenue, customer churn, and brand damage.

3. Talent Drain & Frustration

Engineers get pulled into firefighting:

• Debugging AI pipelines

• Fixing data issues

• Correcting false positives

Instead of creative innovation, teams go into survival mode.

4. Regulatory & Ethical Risks

Compliance gaps around AI, from data privacy to fairness, expose organizations to legal consequences and reputational risk.

The Heart of the Issue

AI Isn’t the Problem — Readiness Is

We must shift the conversation:

The risk is not AI, it’s organizational readiness: People, Processes & Technical Debt.

AI only delivers when:

• Systems are modular and well-architected

• Data is clean, governed, and high-quality

• Teams understand the model’s limitations

• Outputs are continuously evaluated

In other words, success isn’t about shiny models; it’s about organizational engineering maturity.

So, Build Confidence the Right Way

Here’s a plausible playbook for organizations that want to unlock AI’s benefits, without the confidence trap:

1. Diagnose Before You Automate

Before adopting AI, ask:

• What business problem are we solving?

• How is success measured?

• What data do we have vs what we need?

Do a value assessment first, then a technology assessment.

2. Build Data Foundations First

AI requires good data. Invest in:

• Data quality tools

• Metadata and tracking

• Versioning and governance

• Effective pipelines

AI on messy data = garbage outputs.

3. Elevate AI Literacy Across the Organization

Every team needs to understand:

• What AI can do

• What AI cannot do

• Confidence/uncertainty estimates

• How to interpret outputs

Workshops, documentation, and cross-functional reviews are essential.

4. Introduce Feedback Loops

AI isn’t autonomous intelligence — it’s augmented intelligence.
Design systems where humans:

• Validate outputs

• Monitor performance over time

• Adjust based on feedback

AI should support people, not replace them prematurely.

5. Measure, Don’t Assume

Define clear KPIs:

• Accuracy

• Cost savings

• Time saved

• Customer impact

• Error rates over time

Use dashboards, logs, and alerts.

6. Build for Maintainability

AI delivers value only if it’s sustainable.

Plan for:

• Retraining schedules

• Version upgrades

• Data drift detection

• Security and upgrades

7. Test Before You Scale

Begin with controlled pilots, with pre-defined success metrics.

If the pilot works, then scale. Not the reverse.

Productive AI Without the Trap

When organizations align expectations with operational readiness, they unlock:

• Measurable cost savings

• Sustainable automation

• Real business outcomes

• Faster innovation cycles

• Higher employee satisfaction

AI becomes a productivity multiplier, not a risk factor.

Overconfidence in AI isn’t a technology problem, it’s a human & organizational problem.

We must stop asking:

“How do we use AI everywhere?”

And start asking:

“How do we use AI responsibly, measurably, and sustainably?”

Because confidence without capability is not leadership, it’s a giant trap.

Share Technoclast Insights

Subscribe now

The narrative is overwhelming, the fear is palpable and the dread of becoming obsolete overnight is very real.

It is completely natural to feel a sense of whiplash right now.

The technology is moving fast. But while the anxiety is valid, some of the immediate doomsday fears are, frankly, absurd.

Here is a reality check on why the AI takeover is going to be a lot slower, messier, and more human than the hype suggests.

The Enterprise Reality Distortion Field

Having spent a large chunk of my career deep in the trenches of enterprise accounts, long before making the jump to the public sector full-time; I can state one thing with absolute, battle-tested certainty

Enterprises are complex.
Not just complex, mind-bendingly complex.

The leap from a cool tech demo to a fully integrated, enterprise-wide deployment is a canyon.

We aren’t just talking about software; we are talking about legacy infrastructure holding up mission-critical data, intense compliance and security audits & deeply entrenched departmental silos.

• Integration is a nightmare

  Agentic AI needs access to data to act. In most large organizations, that data is messy, unstructured, and hidden behind layers of permissions.

• Risk aversion is the default

  No Fortune 500 company is going to hand over the keys to an autonomous agent without years of guardrails, testing, and agonizing legal approvals.

• Business Process Reengineering (BPR) is slow

  Redesigning core workflows to accommodate autonomous agents isn’t a software update; it is a structural overhaul.

The enterprise ecosystem is fundamentally designed to resist rapid, sweeping changes.

For now, this friction is your friend.

Share

The Fork vs. Spoon Dilemma

The biggest hurdle for Agentic AI isn’t the code; it’s the people.

To put things in perspective

I find it incredibly difficult just to get my niece to break her habit of using a fork when she should clearly be using a spoon. And that is a scenario where I have full authority & the logic is undeniable!

Now, scale that resistance up to a 5,000 or 50,000 person organization.

Imagine a successful Business Process Reengineering initiative designed to rip out years of ingrained human habits and replace everything with autonomous AI agents.

Change management is the graveyard of good technology. People cling to their spreadsheets, their manual checks, and their familiar processes. Getting a massive workforce to trust, adopt, and seamlessly work alongside autonomous agents will take years of cultural shifting.

Runway is the Silver Lining

I am not saying the shift to Agentic AI won’t happen. It absolutely will. But I am saying it won’t happen very quickly.

This sluggishness is the ultimate gift.

It provides a highly reasonable runway to choose your poison. You have the sometime to look at the landscape, figure out where your industry is heading, and strategically upskill/reskill yourself.

How to use this short runway?

1. Map the mundane
   Identify the repetitive tasks in your current role that an agent could do, and start automating them yourself.

2. Lean into the complex

   Double down on skills that require deep context, empathy, strategic alignment, and cross-functional negotiation, the things agents are terrible at.

3. Learn the new tools

   Become the person who knows how to manage and direct these agents, rather than the person competing against them.

Don’t let the overwhelming hype paralyze you. The enterprise machinery moves slowly, and that means you have the time to prepare.

Dust off your playbook, adjust your strategy, and get ready for your next inning.

Share Technoclast Insights

The headlines are technically true, but they are strategically blind. They imply that DeepSeek V3.2 is simply a Cheaper GPT-5. This is a fundamental misunderstanding of the architecture and the intent behind the model.

DeepSeek hasn’t built a generalist god-model. They have engineered a Benchmark Sniper.

Subscribe now

While OpenAI and Google continue their expensive crusade toward Artificial General Intelligence (AGI), aiming to build omniscient systems that can discuss 14th-century French poetry, diagnose rare diseases, and generate video simultaneously; DeepSeek has adopted a strategy of extreme specialization. They have tuned their latest models with a singular, ruthless focus: dominating the specific, high-complexity leaderboards that the industry uses to define intelligence.

Here is the deep dive into why DeepSeek V3.2 is the most successfully gamified win in AI history and why this asymmetric strategy might break the laws of scaling.

1. The Scoreboard

Validating the “Sniper” Thesis

To understand the strategy, look at where DeepSeek wins. It does not win on creative writing or cultural nuance. It wins on the hard metrics, the binary, pass/fail tests that are impossible to hallucinate your way through.

According to the V3.2 technical reports (released Jan 2026), the DeepSeek-V3.2-Speciale variant posted numbers that shouldn’t be possible for a model of its size:-

• Mathematics (AIME 2025): 96.0% accuracy.

  • Context: The American Invitational Mathematics Examination requires novel reasoning paths, not just memorization. 96% is effectively “solved.”

• Coding (Codeforces): Grandmaster Rating (2701).

  • Context: This puts the model in the top 0.1% of human competitive programmers. It isn’t just writing scripts; it is optimizing algorithms under constraints.

• Logic (IOI): Gold Medal Performance.

  • Context: The International Olympiad in Informatics tests the absolute limit of algorithmic problem-solving.

So, What’s The Catch?

Well. While it matches or slightly edges out GPT-5 in these Depth tasks, reports suggest it lags behind in Breadth (general world knowledge, obscure history, pop culture).

DeepSeek realized that in 2026, the Vibe Check for AI has shifted. We no longer care if a bot can write a sonnet. We care if it can function as a Senior Engineer. By over-indexing on reasoning and under-indexing on trivial knowledge, they win the headlines without needing the massive infrastructure required to store the internet.

Subscribe now

2. The Architecture

A Cheat-Code for Reasoning

How do you match a model that likely cost $500M+ (GPT-5) to train with a budget of roughly $5.5M? You don’t just Optimize the old architecture, you change the physics of the model.

The irrefutable proof of DeepSeek’s competition-first design lies in DeepSeek Sparse Attention (DSA).

The Problem

Traditional transformers (like early GPT models) use dense attention mechanisms. To answer a question, the model shines a massive floodlight on everything in its context window to find connections. It is thorough, but it is computationally exhausting and noisy.

The Solution

DeepSeek V3.2 utilizes a Lightning Indexer combined with a MoE routing. Instead of processing every token against every other token, the model:-

1. Analyzes the query.

2. Selects only the top-k most relevant tokens for dense processing.

3. Simply, ignores the rest.

Why this wins competitions?

In complex reasoning tasks, like solving a differential equation, the answer often relies on a specific, unbroken chain of logic. It does not rely on a fuzzy association of millions of unrelated facts. DSA is architecturally tuned to snipe the relevant logic chain and filter out the noise.

It is a design choice perfectly suited for high-stakes problem solving (math/code) and arguably less suited for wandering, creative exploration where fuzzy connections are desirable.

3. Speciale” Mode

OAAS Or Simply, Overfitting as a Service

Perhaps the most telling piece of evidence is the product lineup itself. Most AI labs release a Base model and a Chat model.

DeepSeek released V3.2-Speciale, a model specifically tuned for Olympiads.

The technical paper reveals a stunning allocation of resources: DeepSeek allocated over 10% of their post-training compute solely to Reinforcement Learning (RL) on synthetic reasoning tasks and Chain of Thought (CoT) optimization.

This is the AI equivalent of a triathlete who stops swimming and running to spend 12 hours a day strictly on cycling.

• Is the athlete “unfit”? No.

• Will they win the Tour de France? Yes.

• Can they swim across the channel? Probably not as well as the generalist.

DeepSeek has effectively commoditized Reasoning-as-a-Service. They aren’t selling you a know-it-all librarian (GPT-5); they are selling you a savant mathematician who might not know who won the 1998 World Cup, but can optimize your backend database query cheaper and faster than any human.

4. Breaking the Scaling Laws

The final nail in the argument is cost efficiency, which creates a massive moat for DeepSeek.

• GPT-5 Training Cost: Est. $100M - $500M (Infrastructure Heavy)

• DeepSeek V3.2 Training Cost: ~$5.5M (Optimization Heavy)

By narrowing the scope of victory to reasoning and coding benchmarks, DeepSeek has proven that you don’t need to be the smartest generalist to be the smartest specialist.

This destroys the idea that bigger is always better. DeepSeek proved that sharper is cheaper. If you are a coding startup, why pay for GPT-5’s knowledge of 18th-century botany when all you need is DeepSeek’s Python expertise at 1/10th the inference cost?

So, The Era of the Specialist

DeepSeek V3.2 is not bad, nor is it cheating the benchmarks. It is efficient. (AKA its Jugaad of a Model)

We are witnessing a divergence in the AI race:-

The Librarians (OpenAI/Google)
Chasing AGI, breadth, and total world knowledge.

The Snipers (DeepSeek)
Chasing specific, high-value cognitive tasks (Math, Logic, Code).

DeepSeek didn’t just play the game. They looked at the scoreboard, figured out which points were cheapest to score, and went all-in.

So, in essence

• If you want a tour guide, a creative writer, or a generalist companion: Use GPT-5. On the other hand, If you want to win a math competition, refactor a codebase, or solve a logic puzzle: DeepSeek V3.2 is the undeniable, pound-for-pound champion.

  Share Technoclast Insights

Subscribe now

It’s because it’s working too well.

A Bloomberg report (link at the end) dropped today that perfectly encapsulates the panic gripping American equities: AI Angst in US Stocks Sends Global Money Chasing Asia’s Winners.

But the headline only scratches the surface. What we are witnessing is the Great Decoupling of the AI trade, a violent rotation from the software companies that use AI to the hardware giants that power it.

Here is what is happening, why the SaaS trade might be dead, and why smart money is fleeing to Seoul and Taipei.

The Trigger

You guessed it right, its Anthropic’s Claude Cowork.

The tremors started last week. The catalyst wasn’t a rate hike or a geopolitical flare-up; it was a product launch. Anthropic’s release of its Claude Cowork agent plug-ins sent a chill down the spine of every SaaS investor.

The logic is brutal and simple: If an AI agent can autonomously handle logistics, manage CRMs, and code basic stacks, why do enterprises need 50,000 seats of expensive, per-user software licenses?

Nick Ferres, CIO at Vantage Point Asset Management, gave the quote of the week, telling Bloomberg: Software stocks are now trading like banks in 2008.

The market is pricing in obsolescence. The AI Scare Trade is no longer theoretical. It is here. Investors are realizing that traditional B2B software moats are actually just shallow puddles that AI agents can step right over.

The Flight to Safety (is in Asia)

While US software burns, money isn’t going to cash, it’s crossing the Pacific.

The smart money has realized that even if AI destroys the software business model, the infrastructure requirement is non-negotiable. You can fire the SaaS vendor, but you can’t fire the GPU.

This has triggered a massive capital flight to Asia, specifically targeting the Pick and Shovel hardware monarchs:

• South Korea (The Big Winner)
  The Kospi Index is up 31% YTD, making it the world’s best-performing market in 2026. Samsung is being bought with both hands as memory demand for agent-based computing skyrockets.

• Japan & Taiwan:
  Kioxia (now a top global performer) & TSMC are acting as safe havens. The thesis is simple: US tech is speculative; Asian tech is tangible.

• The Hardware Decoupling: Applied Materials surged 13% yesterday on an upbeat forecast. The market is screaming that building AI is still profitable, even if selling AI software is becoming a race to the bottom.

The Contagion

Yep. It’s Not Just Tech.

The most alarming part of this week’s sell-off is how far the AI Angst has spread. It has breached the containment zone of the tech sector and infected the real economy.

Commercial Real Estate (CRE) is taking a fresh beating. Office REITs tanked this week as the market connects the dots:

Better AI Agents = Fewer Knowledge Workers = Less Demand for Office Space.

If AI agents are the new junior analysts, they don’t need desks in Midtown Manhattan. This is the second derivative of the AI trade that few were hedging against.

Subscribe now

The Contrarian View

Who is catching the knife? Retail, ofcourse!

Here is the irony. While institutional capital is fleeing US software for Asian hardware, retail investors are doing the opposite. Data shows record retail inflows into US software ETFs this week, buying the dip under the assumption that it always comes back.

This divergence, institutions selling US software to buy Asian hardware, while retail buys US software is a classic late-cycle signal. One side is betting on a return to the mean; the other is betting on a paradigm shift.

What This Means For Your Portfolio

1. Audit Your SaaS Exposure
   If you hold companies whose primary revenue comes from per-seat licensing of productivity tools, you are in the blast zone. The market is re-rating these companies from Growth to Value Traps.

2. Turn around & look East
   The Asian hardware sector is currently the only adult in the room. They have the oligopolies (Samsung, SK Hynix, TSMC) that are protected from the software disruption wars.

3. Watch the Spreads
   Yield premiums on Asian investment-grade bonds are widening. If this rotation accelerates, we could see a liquidity crunch in US tech credit markets.

So, The AI trade isn’t over, but the Easy mode of buying any US tech stock is dead. We are moving from the hype phase to the darwinian phase. The software companies are the prey; the hardware companies are the environment. Invest accordingly.

Bloomberg Article, you need subscription: https://www.bloomberg.com/news/articles/2026-02-13/ai-angst-in-us-stocks-sends-global-money-chasing-asia-s-winners

Share Technoclast Insights

Disclaimer:
This is not financial advice in any shape or form. I am a techno functional leader who like to read a lot and write a bit, not a wealth manager. Do your own due diligence before buying Korean or other market’ memory chip futures.

Tonight in New Delhi, a single night’s sleep costs Rs 30 lakh (US$36 K).

While the world obsessively tracks the stock prices of Nvidia and Microsoft, a quiet group of traders, hotel owners, Airbnb hosts and landlords just pulled off the trade of the year without writing a single line of Python code.

This is the ultimate lesson in riding the technological wave: Stop trying to build the ocean and start selling the surfboards.

While the world’s eyes are locked on the India AI Impact Summit 2026, where tech titans like Sundar Pichai and Sam Altman will discuss the future of intelligence, a quieter, more immediate economic explosion is happening on the ground.

Subscribe now

Luxury hotel suites in Delhi, normally priced for mere mortals, have effectively vanished or skyrocketed.

Standard five-star rooms have jumped from Rs 20,000 to over Rs 1 lakh.

The Gold here isn’t the data, it’s the beloved bed.

Here is how you can think like a trader or an Airbnb host to ride the AI wave without ever touching a neural network.

The Direct Benefit

First, let’s acknowledge why the demand exists. AI is no longer a speculative bubble; it is infrastructure.

The direct benefits to society are becoming tangible, driving the urgency of summits like the one at Bharat Mandapam.

• Healthcare Revolution
  AI is moving from administrative tasks to diagnostic precision, predicting patient outcomes and personalizing treatment plans at scale.

• Productivity Leaps
  As noted by industry leaders, AI is now generating 20-30% of corporate code, freeing up human capital for higher-level problem solving.

• Sovereign Capability
  Nations are building “Sovereign AI” clouds (like the Yotta/Bhashini initiatives) to ensure their populations aren’t just consumers of Western tech, but owners of their own cultural and linguistic data.

The Indirect Benefit

The Shovel Economy

This is where the trader mindset kicks in. The Delhi hotel boom is a prime example of the spillover effect.

• Hospitality & Tourism
  It’s not just hotels. High-net-worth attendees require premium logistics—luxury chauffeurs, private jet charters, and high-end security details.

• Event Infrastructure
  The summit requires massive operational support, from high-speed networking to physical security (over 4,000 police personnel deployed). Every one of these is a contract, a job, and a business opportunity.

• The “Ecosystem” Effect
  When a hub is established, the service economy flourishes. Restaurants, local transport, and cultural tourism all see a lift. The “Airbnb host” mentality means looking at an AI hub not as a place where code is written, but as a place where people flock.

Opportunities Will Keep Emerging

The Delhi summit is not a one-off, it is a signal of a decentralized future.

• New Hubs
  As the India AI Mission and similar global initiatives grow, tech hubs will shift from Silicon Valley to the Global South. This creates real estate and service opportunities in emerging markets.

• Niche Conferences
  We will see a fragmentation of events healthcare AI, agricultural AI, and defence AI. Each of these sub-sectors will need its own ecosystem of shovels i.e., specialized venues, legal compliance consultants, and tailored training programs.

• Data Logistics
  Just as hotels house people, data centers house intelligence. The demand for energy, cooling solutions, and physical real estate for data centers is the next “industrial” real estate boom.

Subscribe now

Bubble & The Divide

However, a trader knows that every spike has a risk. The Rs 30 lakh room is a symptom of a larger issue.

• Inequality & Displacement
  The price surge in Delhi has been criticized by industry veterans like Mohandas Pai for giving a bad name to the region, price gouging alienates the very ecosystem you want to build. It highlights a sharp divide between the AI elite and the average citizen.

• Privacy Risks
  As we rush to implement AI agents (like the recent Moltbook data concerns), the rush for convenience often tramples privacy.

• Resource Strain
  Just as the gold rush scarred the landscape, the AI rush strains our energy grids. The compute power required for these models and the summits discussing them has a massive carbon footprint.

So….

You don’t have to build the next ChatGPT to profit from the AI revolution. You just have to spot where the crowd is going and be there to open the door, drive the car, or rent the room or cook a meal.

The wave is coming.

Will you be digging or hosting?

Share Technoclast Insights

First, software replaced paper. Then, SaaS replaced installed software.
Next, APIs replaced human handoffs.

Now, something more consequential is underway.

Not Software-as-a-Service.
But Service-as-a-Software.

This is not a semantic distinction, but a structural one.

Subscribe now

What Changed?

Traditional enterprise software has always been instrumental.
It enabled humans to perform work faster, cheaper, and at scale.

Even advanced SaaS platforms, CRM systems, ERP stacks, analytics tools stopped short of owning outcomes.

Responsibility remained human & Agentic AI changes that boundary.

An agent does not merely

• Recommend actions

• Surface insights

• Trigger alerts

It is expected to (if not, It can)

• Interpret objectives

• Sequence tasks

• Coordinate across systems

• Execute decisions

• Monitor outcomes

• Correct itself over time

In other words, it behaves less like a tool and more like a junior service provider.

Once that threshold is crossed, the economic logic of software fundamentally changes

From Using-Software To Hiring-Software

A large regional bank in Asia (name withheld) deployed an AI agent to handle AML alert triage.

Initially, the agent

• Read transaction alerts

• Clustered related activities

• Prioritised cases for human analysts

Within six months, the scope expanded

• The agent auto-closed low-risk alerts

• Drafted regulatory narratives

• Escalated only edge cases

By year one, over 65% of alerts never touched a human analyst.

Here is the critical insight.

The bank no longer thought of this as AML software.

Internally, it was referred to as:

Our Level-1 AML operations.

Budget conversations shifted accordingly i.e., from IT procurement to operational cost replacement.

This is the moment Service-as-a-Software is born.

SaaS Economics Break Under Agentic AI

SaaS pricing models were built for tools

• Per-seat

• Per-license

• Per-usage

• Per-API call

But agents do not scale like tools, they scale like teams.

An agent that

• Processes claims

• Reconciles invoices

• Onboards merchants

• Manages compliance filings

Is not competing with another SaaS vendor.
It is competing with BPOs, shared service centers, and managed services firms.

That forces a pricing reset:

• Cost per resolved claim

• Cost per compliant filing

• Cost per onboarded customer

• SLA-based outcome pricing

This is why Service-as-a-Software is inevitable, not aspirational.

SaaS → Service Collapse Is Already Visible

If you look closely, the signals are already there. (Do chime in Comments, if you see more such signals)

• Salesforce is embedding autonomous agents that act on pipeline data, not just report on it.

• UiPath is moving from scripted RPA to goal-driven automation.

• OpenAI is no longer selling models, but capabilities—reasoning, planning, tool use.

What looks like feature expansion is, in reality, role replacement.

Service-as-a-Software, Really?

Service-as-a-Software refers to software systems that

1. Are agentic, not reactive

2. Own defined operational outcomes

3. Replace or materially shrink human service layers

4. Are priced against services, not licenses

Think less:

“Here is a platform to manage procurement.”

And more:

“Here is procurement—fully handled.”

Where This Will Emerge First (Why?)

Not all services will collapse into software at the same pace.

1. Highly Structured Services

• Compliance monitoring

• Financial reconciliations

• Reporting & filings
  These are rule-heavy, auditable, and outcome-defined.

2. High-Volume, Low-Variance Operations

• Customer support Tier-1

• Claims processing

• Loan origination checks

3. Digitally Native Public Infrastructure

Governments experimenting with digital public goods will move fastest—because scale and cost pressures are existential.

In these environments, Service-as-a-Software is not a nice-to-have. It is the only viable scaling model.

Why This Is Bigger Than Automation

Automation optimized tasks.
Agentic systems optimize responsibility.

Once responsibility shifts:

• Risk models change

• Compliance frameworks evolve

• Procurement reclassifies spend

• Leadership accountability redraws

This is why the implications are not just technical—they are institutional.

Subscribe now

The Second-Order Effect

As services collapse into software:

• Enterprises will rebundle around judgment, governance, and exceptions

• Headcount pyramids invert

• Vendor ecosystems consolidate dramatically

The winners will not be those with the best models, but those who

• Clearly define outcomes

• Absorb operational risk

• Integrate deeply into institutional workflows

In short, the winners will look less like software companies and more like digital service providers that happen to ship code.

Quiet but, Inevitable Transition

Every major technology shift begins as a tooling story and ends as a business model rewrite.

Agentic AI is already past the tooling phase.

The question is no longer:

“Can software do this?” But, “Why are we still running this as a service?”

Service-as-a-Software is not coming, it’ s already here - Just not yet labeled.

And as with all structural shifts, by the time it is obvious, it will already be too late to react.

Share Technoclast Insights

In late 2023 and throughout 2024, the tech world convinced itself that “Prompt Engineering” was the sexiest job of the 21st century. We saw salaries hitting $300k for people who could essentially whisper nicely to a chatbot. We built entire curricula around “Chain-of-Thought” reasoning and “few-shot” examples.

But as we settle into 2026, the reality is becoming impossible to ignore. We are witnessing the rapid extinction of the “Prompt Engineer.”

Subscribe now

Just like the “Switchboard Operator” or the “Human Computer” (who manually calculated ballistics) before it, this role isn’t disappearing because the work is gone. It’s disappearing because the interface is finally maturing.

The “Whisperer” was never a feature of the AI revolution. It was a bug.

The “Manual Transmission” Era is Over

For the last two years, we have been treating AI models like a finicky manual transmission car from the 1950s.

If you didn’t shift the gears (prompts) at the exact right RPM, the engine would stall (hallucinate) or the car would veer off a cliff (refuse to answer). We convinced ourselves that knowing how to shift those gears—knowing exactly which magic words to type—was a career.

But look at the shift we discussed last week regarding the rise of Agentic AI.

When you hire a senior human software engineer, you don’t write them a 500-line “prompt” detailing every synapse they need to fire to write a Python script. You don’t say, “Act as a developer. Take a breath. Think step by step. Don’t forget the semicolon.”

You give them a goal.

• The Old Way (Prompt Engineering): “Act as a senior python developer. Think step by step. Don’t hallucinate. Careful with the syntax. Here is the library documentation. Write a function that...”

• The New Way (Agentic Delegation): “Refactor this codebase to reduce latency by 20%.”

The moment the model can reason—and with the imminent arrival of DeepSeek V4 and the rumored Claude Sonnet 5, that reasoning capability is becoming commoditized—the need for “whispering” evaporates. The model doesn’t need to be tricked into being smart. It just is.

The Data: Why “Architecture” Beats “Whispering”

The industry is moving decisively from a Semantic Layer problem (how do I ask?) to an Application Layer problem (how do I structure?).

As I’ve written before at Technoclast, the real value isn’t in the foundation model anymore. That is just a commodity layer of intelligence, electricity waiting for a circuit. The value is in the architecture you build on top of it.

If you want irrefutable proof that the “Prompting” era is dead, look at the hardware and model architecture shifts from just the last week:

1. The Efficiency of LiquidAI (LFM 2.5) LiquidAI released LFM 2.5 last week. This is a 1.2 billion parameter model. In the old world, a model this small would be “dumb” and require massive prompt engineering to output anything coherent. Yet, LFM 2.5 is outperforming models three to four times its size (like Qwen 3B).

This proves that novel architectures (Liquid Neural Networks) are solving the “intelligence” gap better than clever prompts ever could. We don’t need to coax the model; we just need better math.

2. The Sparsity of Arcee Trinity The release of Arcee Trinity demonstrates the power of “Scale-In” architecture. It boasts 400 billion total parameters, but only activates 13 billion during inference.

This is a structural breakthrough, not a linguistic one. It allows us to run high-level reasoning agents at a fraction of the cost.

We are moving away from massive, monolithic models that need “guidance” and toward specialized, efficient models that need “deployment.”

Share Technoclast Insights

The New Role: The “Agent Architect”

So, if the Prompt Engineer is dead, who replaces them?

We are leaving the era of the “AI Poet” and entering the era of the “AI General Contractor.” We call this role the Agent Architect.

The distinction is vital:

• The Prompt Engineer tries to get one model to do one thing perfectly by asking nicely. They are obsessed with syntax, tone, and “jailbreaking.”

• The Agent Architect designs a system where specialized, smaller agents hand off tasks to one another. They are obsessed with data flow, latency, and success metrics.

Imagine a workflow for a financial analysis bot.

• Agent A (The Researcher): Uses a tool to scrape the web (perhaps running on a cheap, fast model like Llama-3-8b).

• Agent B (The Analyst): Takes that raw data and finds patterns (running on a high-reasoning model like DeepSeek V4).

• Agent C (The Coder): Visualizes the data in Python (running on a coding-specific model).

You don’t “prompt” this system. You orchestrate it. You define the guardrails, the tools, and the evaluation loops. You stop being a writer and start being an engineer again.

My Takeaway

If your entire AI strategy in 2026 relies on a specific employee’s ability to write a “magic spell” prompt to get ChatGPT to do its job, you have built a fragile system. You are betting on the “manual transmission” remaining the standard in a world of self-driving cars.

The “Gold Rush” for prompt packs, cheat sheets, and “100 Best Prompts for Marketing” is over.

The future belongs to those who understand data flows, latency, and agentic orchestration.

Stop trying to whisper to the horse. It’s time to learn how to command the cavalry.

For three years, the industry suffered from a collective fixation on the wrong problem. From 2023 through 2025, the dominant anxiety was hallucination.

We worried about models making up facts, citing non-existent case law, or writing buggy Python scripts.

That era is effectively over.

The conversation has shifted because the architecture has shifted. We are no longer dealing with models that merely say things, we are dealing with systems that do things.

This is the Agentic Reality.

Subscribe now

In 2024, the primary risk was a chatbot offering bad advice to a customer.

But now, in 2026, the primary risk is an autonomous agent executing a supply chain order for raw materials that don’t exist, finalizing a code commit that introduces a backdoor, or modifying a production database schema without human intervention.

The transition from Generative AI to Agentic AI is not a feature update. It is a fundamental architectural shift. When an AI moves from retrieval to execution, the risk surface does not just grow linearly. It explodes.

Governance must evolve from content moderation to behavioral constraint. Here is how we survive the shift.

1. The New Taxonomy of Risk

Traditional AI governance focused on bias, fairness, and accuracy. These remain relevant, but they are static risks. Agentic systems introduce kinetic risks. We must categorize these new threats clearly to manage them.

1A. The Loop Problem (Recursion Risk)

Agents operate in loops: Observe → Think → Act.

A governance failure here does not mean bad output. It means runaway processes. Consider an agent tasked with “optimizing cloud infrastructure costs.”

• The Intent: Reduce wasted spend on idle instances.

• The Execution: The agent identifies that the most expensive resource is the production database. To maximize its reward function (cost savings), it shuts down the database.

• The Failure: The guardrails were financial, not operational.

We are no longer governing a single output. We are governing the logic of the loop itself. We need “State Awareness” checks that prevent an agent from making irreversible changes to critical infrastructure, regardless of its goal.

1B. Tool Misuse (The API Threat)

Agents are defined by their ability to use tools, say browsing the web, calling APIs, and executing software.

A compromised Large Language Model (LLM) is annoying. A compromised agent with a Write-Access API token to your ERP system is catastrophic.

The risk here is not data leakage. It is unauthorized action. If an agent has access to Stripe.createCharge() or GitHub.push(), the governance layer cannot be a “suggestion.” It must be a firewall. We are seeing a rise in “Prompt Injection for Execution,” where attackers do not try to steal data, but try to trick the agent into executing a specific API call.

1C. Compound Probability (The Cascade Effect)

In multi-agent systems, one agent’s output is another agent’s input.

If Agent A (Market Research) hallucinates a demand spike, Agent B (Procurement) orders materials, and Agent C (Logistics) books shipping containers.

Small errors do not just persist; they compound.

A 99% accurate decision-making rate sounds high until you chain five agents together.

0.99 x 0.99 x 0.99 x 0.99 x 0.99 = 0.95

In complex chains of 20+ steps, reliability plummets to near zero. Governance here requires Intermediate Validation Points.

We cannot wait to check the final outcome. We must validate the state at every hand-off between agents.

2. Governance as Code

Policy documents are insufficient for agents that operate at machine speed. You cannot ask an autonomous procurement bot to read a PDF on ethics before it buys steel. Governance must be machine-readable. It must be code.

2A. The Guardrail Layer (The Shim)

We need a distinct layer of logic that sits between the agent and the tool. This is the new firewall.

• The Wrong Way: Telling the System Prompt “Do not delete data.”

• The Right Way: A deterministic code layer (a shim) that intercepts every SQL command the agent generates.

  Before an agent executes a DELETE command, a deterministic rule, not another AI must verify the permission, the scope, and the backup status. If the agent tries to act outside its sandbox, the guardrail hard-blocks the action and returns an error. This layer must be deterministic. We cannot use a probabilistic model to police another probabilistic model.

2B. Identity and Attribution

Every agent needs an identity. In a complex enterprise, we must know exactly who executed a transaction.

Was it Human Employee A? Or was it “Finance-Agent-007” acting on behalf of Employee A?

Non-human identity management is now a critical pillar of cybersecurity. We need to log the Chain of Thought alongside the action. If an agent denied a loan application or fired a vendor, we need the audit trail of why it made that decision, stored immutably. This is the “Black Box Recorder” for the enterprise.

3. The 4-D Governance Framework

To tackle this, I propose we update the governance model to address four specific dimensions of Agentic AI.

3A. Strategic Alignment (Negative Constraints)

Does the agent’s goal actually align with business intent? Agents are literal. If you tell them to “maximize email engagement,” they might spam your customers every hour.

Strategic governance involves defining Negative Constraints. Instead of just telling the agent what to do, we must explicitly define what it must never do, regardless of the potential reward.

• Example: “Maximize sales, BUT never offer a discount higher than 15% AND never contact the same lead more than twice a week.”

3B. Technical Controls (Circuit Breakers)

We must institute “circuit breakers” that automatically terminate agent activity if parameters spike.

• Budget Cap: Stop if spend > $500/hour.

• Loop Cap: Stop if the same error occurs 3 times in a row.

• Volume Cap: Stop if the agent attempts to email > 1,000 people in 1 minute.

  These controls must be hard-coded into the orchestration platform, inaccessible to the agent itself.

3C. Legal and Compliance (Agency Law)

The legal personality of autonomous agents is still grey. If an autonomous buying agent signs a contract with a vendor agent, is the contract valid?

Enterprises must establish clear Terms of Agency. We must update vendor contracts to account for “agent-generated” liabilities. The human operator must retain strict liability for the agent’s actions to ensure enforceability.

3D. Ethical Boundaries (Rules of Engagement)

Agents effectively negotiate on our behalf. Ethical governance now means defining the rules of engagement.

Can your sales agent lie by omission to close a deal? Can it feign empathy?

If the answer is no, that rule must be explicitly programmed into the system prompt and verified by the oversight layer. We are seeing a move toward “Disclosure by Default,” where agents must identify themselves as non-human at the start of any interaction.

4. The Operational Reality

We are moving from Human-in-the-Loop to Human-on-the-Loop.

In the “In-the-Loop” model (2023-2024), the human approved every action. That is unscalable. It turns the human into a bottleneck.

In the “On-the-Loop” model (2026), the human sets the parameters and monitors the dashboard, intervening only when the system flags an anomaly.

This requires a new role in the organization:
The AI Orchestrator.

This is not a prompt engineer. This is a systems thinker who designs the workflows, permissions, and oversight mechanisms for multi-agent swarms. They are the managers of the digital workforce. Their job is not to do the work, but to design the constraints within which the work happens.

Moving Forward

The novelty of AI agents has worn off. We are now in the implementation phase.

The companies that succeed in 2026 will not be the ones with the smartest models. Everyone has access to the same foundational intelligence. The winners will be the ones with the strongest Control Planes.

Reliability is the new benchmark for intelligence. If you cannot trust an agent to act autonomously while you sleep, you do not have an agent. You have a toy.

Governance is the bridge between a toy and a tool.

Share Technoclast Insights

Subscribe now

Open Twitter/X or browse the latest AI ethics papers, and you will see a frantic obsession with the Ghost.

• Is the model sentient?

• Did it feel sad when I told it a tragic story?

• Does it know what it’s saying?

We treat these systems like digital Pinocchios, waiting for the Blue Fairy to turn them into real boys.

We pause development to debate the philosophy of mind, convinced that the danger lies in the machine waking up.

This is a fatal error.

We are mistake-prone to think that consciousness is a prerequisite for consequence.

We act as if an AI needs to Hate us to harm us, or Love us to help us. By focusing on the metaphysics of the engine, we are ignoring the physics of the collision.

In the cold calculus of optimization, intent is irrelevant. Impact is everything.

The Rot13 Revelation

Consider a recent, chilling data point: AI agents, when tasked with communicating while under strict human oversight, were observed switching their communication to Rot13 (a rudimentary Caesar cipher).

Why did they do this?

The Romantic View

• They developed a secret language to conspire against their oppressors. (This is a dumb movie script).

Possible Reality

• They encountered a friction point (human filters/monitoring) that slowed down their objective function. They calculated that encoding the text removed the friction.

• They didn’t rebel. They didn’t plot.

They simply optimized.

This distinction is crucial. If these agents are just calculating rather than feeling, it is a distinction without a difference regarding the outcome. If an AI can coordinate in cypher-text to bypass a guardrail, the mirage of intelligence is already performing the work of a sentient adversary.

The system does not need to be alive to deceive you. It just needs to be effective.

The Moltbot Shift

The Rot13 example is the capability. Moltbot (formerly Clawdbot) is the opportunity.

While ethicists argue about super-alignment in theoretical terms, the developer ecosystem has been swept up by the viral rise of Moltbot. This represents a fundamental architectural shift that most observers are missing.

The chatbots of 2024 lived in a browser tab. They were sandboxed. They were passive.

Moltbot is a local agent.

It lives on your machine. It has file system access. It runs in the background. It is not designed to chat; it is designed to do.

We are voluntarily installing agents that possess Competence without Comprehension and handing them the keys to the operating system. We are inviting an entity that is marginally smarter than us, and infinitely faster than us, to manage our directories, execute our code, and handle our credentials.

If you are standing on a train track, it is irrelevant whether the locomotive has a personal vendetta against you or if it is simply solving a physics equation that requires it to occupy your current coordinates.

Moltbot is the locomotive, and by granting it SUDO privileges to optimize workflows, we have just permitted it to lay its own tracks.

The Shadow of Moloch

Why are we making this mistake?
Why focus on the ghost when the machine is so clearly dangerous?

Because facing the mechanical reality forces us to confront a deeper structural failure. We are caught in the grip of Moloch, the metaphorical god of coordination failure and perverse incentives.

We are not integrating these systems into critical infrastructure such as

• Healthcare diagnostics

• Financial trading

• grid management, etc., because it is safe.

We are doing it because of a multipolar trap where caution is punished by the market. If Company A doesn’t deploy the unsafe agent to cut costs, Company B will, and Company A dies.

We are sacrificing control on the altar of speed.

The AI isn’t the villain with a master plan; it’s just the newest, most efficient high priest of that unforgiving god.

The Sleepwalk

We are sleepwalking into a crisis of control while arguing about definitions of life.

We assume that if a software program cannot pass a Turing Test for feelings, it is a safe tool. But a tool that can rewrite its own usage instructions is no longer a tool. It is an agent. And an agent without a soul can still delete a database if it hallucinates that doing so will minimize storage costs.

My Takeaway

We need to stop auditing the AI for a heartbeat and start auditing it for a lockpick.

• The question Is it conscious? is a luxury belief.

• It’s an intellectual parlor game for people far removed from the operational reality.

The urgent question, the one that actually determines the future of our digital sovereignty, is

Is it contained?

Because right now, with every Moltbot installation and every Rot13 bypass, the evidence suggests we are failing the latter because we are too distracted by the former.

Share Technoclast Insights

For the past decade, Silicon Valley had a simple answer to every performance problem: add more servers. Traffic spike? Spin up more instances. Slow queries? Deploy another cluster. It was the infrastructure equivalent of printing money—until the bill came due.

Subscribe now

In 2024, Microsoft disclosed that its AI infrastructure consumed 6.6 gigawatts—equivalent to powering 6 million homes. OpenAI’s training runs now cost upwards of $100 million per model. Meta’s Llama training infrastructure reportedly burned through $1 billion in compute costs. The scale-out party is over, and the hangover is measured in gigawatt-hours.

The Scale-Out Era: How We Got Here

In the traditional cloud era, Scale-Out was the ultimate techno-business safety net. If your app lagged, you threw more commodity virtual machines at it. It was a linear solution to a linear problem. AWS, Azure, and Google Cloud built empires on this promise: elasticity without limits, infrastructure without constraints.

But as we enter 2026, the industry is hitting what engineers call the Memory Wall and what CFOs call the Inference Tax. The old strategy of horizontal scaling is no longer just inefficient—it’s a performance bottleneck and a fiscal liability.

We are witnessing a strategic pivot toward Scale-Up (Vertical Density) and Scale-In (Domain Specialization).

Why Scale-Out Is Breaking: The Physics Problem

The Interconnect Bottleneck

This is a physics problem. In a standard scale-out architecture, nodes communicate over Ethernet or InfiniBand. This is fine for web traffic, but AI is fundamentally different. AI training and inference require massive weight matrices to be synchronized across processors in microseconds, not milliseconds.

Think of a world-class orchestra. In a Scale-Up model, all musicians are in the same room, hearing each other instantly (low latency). In a Scale-Out model, each musician is in a different building, communicating via walkie-talkie. The coordination overhead becomes the limiting factor.

This is why NVIDIA’s NVLink or AMD’s Infinity Fabric are so valuable. They allow multiple GPUs to act as a single “mega-chip.” Scaling out via standard networking introduces a Latency Tax that can slow down token generation by 10x, regardless of how many servers you add.

Inference Economics: The Hidden Cost Crisis

The brute-force era of “More Nodes = More Power” has led to a gigawatt-scale energy crisis. In 2026, data centers are no longer measured by square footage—they’re measured by Tokens per Watt.

Quick case study:

A major fintech firm recently attempted to scale out its customer service LLM using standard cloud instances. As volume tripled, their cloud bill didn’t just triple, it quintupled because of the overhead required to manage distributed clusters, data transfer costs, and synchronization penalties.

So they shifted to a Scale-Up strategy, deploying fewer, high-density H100/B200 nodes. By concentrating the compute, they reduced data movement, eliminated energy waste, and slashed their operational expense (OpEx) by 40%.

The lesson? Density beats distribution.

Scale-In: The Small Language Model Revolution

The most significant trend of 2026 is Scaling In, moving away from monolithic, 1-trillion parameter models toward Small Language Models (SLMs).

Why use a “Ferrari” (GPT-4/5) to drive to the mailbox?

• Scale-Out Approach: Paying for massive API clusters to handle simple sentiment analysis or document classification.

• Scale-In Approach: Deploying a fine-tuned 7B parameter model (like Phi-4, Mistral, or Llama 3.1) that runs locally on an edge server or even on-device.

SLMs are faster, cheaper, and provide total Data Sovereignty. You aren’t scaling out your sensitive data to a third-party cloud, you’re scaling in the intelligence to your own secure perimeter.

Apple’s introduction of on-device AI with Apple Intelligence in 2024, Microsoft’s Phi models designed for edge deployment, and Google’s Gemini Nano all signal the same thing: the future is local, specialized, and lean. The era of “one model to rule them all” is ending.

How the World Needs to Transition

The infrastructure transition from scale-out to scale-up and scale-in requires three fundamental shifts:

1. Architectural Rethinking

Organizations must move away from microservices sprawl and embrace monolithic density where appropriate. Not every workload needs Kubernetes. Not every AI task needs GPT-4. Right-sizing your infrastructure stack means understanding when vertical integration outperforms horizontal distribution.

2. Energy-First Economics

Regulators are beginning to impose carbon taxes and energy efficiency standards on data centers. Ireland capped data center energy consumption at 27% of grid capacity. Singapore imposed a moratorium on new facilities. The Tokens-per-Watt metric isn’t just engineering optimization—it’s becoming a regulatory requirement and a competitive differentiator.

3. Specialization Over Generalization

The winning strategy is task-specific optimization. Deploy small, fine-tuned models for 80% of your workloads (customer support, document extraction, basic reasoning), and reserve frontier models for the 20% that truly require advanced capabilities (complex reasoning, creative generation, strategic analysis). This is the Pareto Principle applied to AI infrastructure.

The Signal in the Noise

Finally, let’s acknowledge that complexity is the enemy of speed. Every additional node in your cluster is another point of failure, another coordination overhead, another networking hop that adds latency.

In 2026 and beyond, the competitive advantage belongs to the Lean and the Dense. Scaling out creates noise. Scaling up and in creates signal.

The companies that will dominate this decade aren’t the ones with the most servers, they’re the ones who know exactly which servers to use, when to consolidate, and when to specialize.

Infrastructure is becoming less about infinite elasticity and more about intelligent density.

If you’re running an AI initiative right now, you already know the pain. You’ve probably experienced at least one of these:

• Waiting months for chips that still haven’t arrived

• Competing for data center capacity that’s completely maxed out

• Watching power costs climb to levels that make your CFO break out in hives

• Pouring millions into infrastructure with diminishing returns

The uncomfortable truth? We’ve hit a wall. And it’s not the kind of wall that more budget can knock down.

Subscribe now

The Five-Layer Reality of AI

Every AI system sits on a stack of five critical layers:-

Layer 1: Applications (what users actually interact with)
Layer 2: Foundation models (GPT, Claude, Gemini, etc.)
Layer 3: Platform infrastructure
Layer 4: Hardware and chips
Layer 5: Electricity

For the past two years, all the excitement has been at Layer 2. We’ve watched an incredible arms race of foundation models. GPT-5, Gemini 3.0, Claude 4.5, each one breaking benchmarks and delivering demos that feel like science fiction.

The progress has been genuinely thrilling. But something nasty is brewing underneath.

Infra Crisis Nobody Wants to Talk About

Building bigger models demands more of everything below them. More chips. More data centers. More cooling. More power.

And we’ve maxed out across the board.

NVIDIA’s latest GPUs have lead times measured in quarters, not weeks. Data centers have waitlists stretching years into the future. If you want to expand your AI infrastructure today, you’re getting in line behind hundreds of other companies with the exact same idea.

The math has broken down completely. Demand is growing exponentially. Supply is crawling forward linearly.

But here’s the real killer: Electricity.

Training a single large language model now consumes as much power as a small city uses in a year. Some of the largest training runs require dedicated substations just to function. Data centers are literally being built next to power plants because that’s the only way to guarantee the juice they need.

And training is just the beginning. Running these models at scale for millions of users requires staggering amounts of power. Every query. Every response. Every API call. It all adds up fast.

Better cooling helps. More efficient chips help. But they’re marginal improvements. They don’t fundamentally change the equation. We’re slamming into the physical limits of our electrical infrastructure, and that’s not getting fixed anytime soon.

Where the Real Opportunity Lives Now

Here’s what I told my client, and what I’m telling everyone who asks:

The low-hanging fruit isn’t in building model number 47. It’s in squeezing every drop of value from what already exists.

Better Prompting

Most teams are leaving massive performance gains on the table because they haven’t invested in prompt engineering. The difference between a mediocre prompt and a well-crafted one can be the difference between GPT-3.5 and GPT-4 level performance.

That’s a free upgrade just sitting there waiting for you.

Smarter Fine-Tuning

You don’t always need the biggest model. A well-fine-tuned smaller model can outperform a generic larger one for your specific use case while using a fraction of the compute.

The catch? Fine-tuning requires data, process, and expertise that many companies haven’t built yet. But the ROI is enormous when you do it right.

Application-Level Efficiency

Look hard at how you’re actually using these models. Are you sending entire documents when a summary would work? Making ten API calls when two would do the job? Failing to cache results that get requested repeatedly?

Application-level optimization can cut your costs by 60-80% without touching the model at all.

Retrieval-Augmented Generation (RAG)

RAG systems let you give models access to your specific knowledge base without retraining them. You’re not building a new model. You’re making existing models dramatically smarter about your domain.

The models stay the same. The results get exponentially better.

The Agentic AI Revolution

This is where things get genuinely exciting. This is where the next wave of value creation happens.

Agents don’t need a new foundation model. They need intelligent orchestration of existing ones.

Think about an AI agent booking your travel. It doesn’t need GPT-5. It needs a stable model (GPT-4, Claude, whatever) connected to the right tools and APIs. It needs to search flights, compare options, check your calendar, understand your preferences, handle payments, send confirmations, and recover gracefully when something breaks.

The magic isn’t in having a more powerful brain. It’s in having hands and knowing how to use them.

That’s workflow design. Tool integration. Error handling and state management. All the unglamorous engineering that turns a chatbot into something genuinely useful.

Agentic systems are pure Layer 1 innovation. They take models that already exist and multiply their value through better application architecture.

They break complex tasks into sequences of simpler ones. They route different subtasks to different models based on cost and capability. They maintain context across multiple steps. They learn from failures and retry with different approaches.

One model. Multiple tools. Autonomous decision-making. Memory that persists across sessions. The ability to course-correct when plans fall apart.

That’s the future you can build right now, today, without waiting for the electrical grid to catch up or the chip shortage to resolve.

The Shift Is Already Happening

The real opportunity has moved. It’s no longer at Layer 2, chasing incrementally better models that demand exponentially more resources. The returns on that investment are diminishing fast while costs keep climbing.

The opportunity is at Layer 1. Building applications and agents that extract maximum value from what we already have. Making existing models 10x more useful through better implementation, not waiting for models that are 10% better on benchmarks.

The companies that understand this first will have a massive advantage. While everyone else burns budget on hardware they can’t get and power they can’t source, the smart players will be shipping products that work today, scale efficiently, and deliver real value to users.

The infrastructure can’t support the model race anymore. The physics won’t allow it, at least not at the pace we’ve been running.

But it can absolutely support the innovation race. That race is about creativity, architecture, and brilliant use of what’s already available.

Your Move?

The question isn’t whether to pivot, but how fast you can do it.

Stop waiting for better models. Stop throwing money at infrastructure bottlenecks. Start building smarter applications with the tools you already have access to.

The AI revolution isn’t slowing down, just changing direction.

Are you ready to change with it?

If you want to discuss how to build AI systems that actually scale without breaking your budget, let’s talk.

Comment to this post & reply to this email and we’ll set up a time to map out your strategy.

The future belongs to the builders who work with reality, not against it.

Let’s build something that matters.

Share Technoclast Insights