Cognition Is Not Construction: Why AI Will Not Replace Physical Work

In recent months, a curious doom narrative is being propagated, perhaps inadvertently, by celebrated tech leaders. This narrative basically extrapolates AI's success in cognitive domains to the physical economy, concluding that AI will soon replace physical human labour.

Two recent, prominent examples illustrate this drift particularly well.

First is Eric Schmidt. At a media event recently, he highlighted the 'danger' of AI with a 'possibility': He just has to tell an AI he wants to build a house in his beloved Virgina state, and AI can take over from there: AI can select the best location, design the house, hire the building company, negotiate the price, write the contract, and later can even falsely sue the builder for 'violating' the contract! 

The other is Sridhar Vembu. At the India Economic Conclave, he noted, rightly, that AI is rapidly displacing coding and routine software jobs. But he went on to predict that a similar fate is awaiting deep-tech and physical industries too, because of potential mass-scale usage of robots.

At first glance, these claims sound plausible—especially to audiences immersed in software-centric narratives. On closer inspection, however, they rest on a fundamental category error.

The Core Mistake: Confusing Coordination With Production

What AI is exceptionally good at today is coordination:

searching,

planning,

drafting,

simulating,

negotiating,

documenting,

optimising workflows.

What AI is not good at is instantiating physical reality.

A house is not built by contracts.

An aircraft is not manufactured by documentation.

A semiconductor fab does not function on spreadsheets and code alone.

They are built by trained human beings working with matter, under conditions of uncertainty, variability, and irreversible physical consequences.

When Schmidt imagines an AI “building” a house, he is actually describing an AI managing paperwork and procedures around a house that is still built, brick by brick, wire by wire, pipe by pipe, by human workers. His example inadvertently proves the opposite of what it intends: that even the most “AI-heavy” scenario remains fully embedded in human labour, institutions, and enforcement.

Why Software Jobs Are Vulnerable—and Physical Jobs Are Not

Vembu is right in his assessment that coding and routine software work are under severe pressure from AI. But this truth should not be used to generalize across industries. 

Software is uniquely exposed because it is:

symbolic,

fully digital,

cheap to iterate,

easy to copy,

and executed in simulated environments.

AI thrives here because cognition → text → execution forms a closed loop.

Physical industries do not work that way.

They are:

embodied,

stochastic,

safety-critical,

site-specific,

regulated,

and full of edge cases that cannot be abstracted away.

The moment something breaks in the physical world, there is no “undo” button. A cracked turbine blade, a misaligned rail track, a contaminated pharmaceutical batch, or a flawed semiconductor tool is not a software bug—it is a costly, often dangerous failure that requires human judgment and accountability.

The Industries AI Cannot Simply “Replace”

Consider just a partial list of sectors often waved away in AI doom narratives:

Aircraft manufacturing and MRO

Shipbuilding and ship repair

Rail manufacturing and maintenance

Semiconductor fabs and fab-equipment manufacturing

Quantum hardware

Space-tech hardware

Energy infrastructure

Mining and refining

Industrial materials and components

Pharmaceutical materials

Public infrastructure construction

Real estate construction

Robotics materials and components (ironically enough)

Every one of these domains relies on trained human workers, tacit knowledge, field improvisation, certification regimes, and physical accountability.

AI can enhance productivity here—through diagnostics, design assistance, predictive maintenance, and planning—but it does not replace the worker. The loop never closes without human hands and human responsibility.

If AI could truly replace these jobs, companies like TSMC, ASML, Boeing, Airbus, or SpaceX would already be doing so at scale. They are not—not because they are sentimental about labour, but because physics, economics, and risk make such replacement infeasible.

Where Robotics Actually Fits

There is a role for AI-powered robots—but it is narrower and more prosaic than doomers suggest. Robotics works best in:

standardised,

repetitive,

high-volume,

low-variance tasks,

typically at the assembly stage.

This is not new. It is exactly how automation has always entered industry. What robots do not replace is:

field judgment,

adaptive problem-solving,

safety sign-offs,

craft skills,

or responsibility when things go wrong.

Deep-tech industries, especially frontier ones like quantum or space hardware, are often pre-standardisation. They are still experimenting, redesigning, failing, and learning. Automation comes after stabilisation, not before.

The Sociological Blind-Spot of Tech Elites

Why do intelligent people like Schmidt and Vembu miss this?

Because their professional lives unfold in worlds where:

abstraction dominates,

paperwork feels like reality,

scaling is frictionless,

failure is cheap,

and physical constraints are invisible.

From that vantage point, it is easy to mistake symbolic power for material power.

But civilisation is not software.

It is cognition embedded in matter.

Until AI can autonomously:

mine materials,

fabricate components,

build factories,

repair itself,

power itself,

and operate safely in chaotic real-world environments,

it will remain what it is today: a powerful brain without a body.

Why This Narrative Is Dangerous

The real danger is not AI replacing physical workers tomorrow. The danger is: policymakers believing it will. And this may result in:

vocational skilling being deprioritised,

industrial labour being underinvested,

and societies sleepwalking into shortages of real capabilities.

For countries like India, this matters enormously. The future lies not in shrinking physical capacity, but in expanding it—using AI as a multiplier, not a mirage.

Conclusion

AI replaces symbols with ease.

It assists work involving matter far more gradually.

It does not displace an embodied civilisation merely by thinking harder.

Coding roles are vulnerable precisely because software is, at its core, pure cognition.

Deep-tech roles endure because they are cognition inseparably anchored in the physical world.

The sooner we stop overflying these details, the healthier our industrial, educational, and policy decisions will be.