The Age of
Abundant
Intelligence

Eighteen ninety-five. Manchester, England. A textile mill owner has just installed his first electric motor. A salesman has convinced him it is the future. So he replaces his steam engine with the electric motor — bolts it onto the same drive shaft that ran the old engine. His factory layout is unchanged. His workflows are unchanged. His management is unchanged. He measures his productivity. The gain is — modest. Disappointing, even.

What that mill owner did not know was that the real productivity gains from electricity would not arrive for thirty more years. Not until factories were completely redesigned around the new power source. The bolt-on approach yielded a few percentage points. The redesign approach yielded ten times the output.

I tell that story because right now, in 2026, we are the textile mill owner. And the electric motor is artificial intelligence.

The thinker behind this episode

This entire essay leans on the work of one researcher: Karan Girotra at Cornell Tech. Charles H. Dyson Family Professor of Management. One of the founding faculty of Cornell Tech. His current research is specifically on how AI automates and augments cognitive work to radically improve productivity. He is not a consultant. He is not a journalist. He is an empirical researcher who runs controlled experiments.

The steam engine analogy I just opened with — that is his core organisational insight. He said it most clearly at the Cornell Executive MBA Summit in March 2025:

The missing piece in the AI revolution isn't smarter AI. It's leaders who know how to restructure their businesses around it.

Most organisations are bolting AI onto existing workflows. They are getting modest gains. The transformational gains will come — but only when the entire process is redesigned.

The seventy-year productivity arc

Every wave promised a productivity revolution. The pattern matters.

First. The mainframe era, 1950s through the 70s. Real productivity gains, but concentrated in back-office functions. The nature of knowledge work did not change.

Second. The PC revolution, 1980s and 90s. Robert Solow wrote: you can see the computer age everywhere except in the productivity statistics. They took fifteen years to show up.

This is the Solow Productivity Paradox. Technology disappoints initially because productivity requires complementary investments. New skills. New processes. New organisational structures. The technology is the easy part. The redesign is the hard part.

Third. Internet and ERP, mid-90s to mid-2000s. US productivity surged. Email created the first attention-fragmentation crisis.

Fourth. Smartphones, cloud, SaaS. 2007 onwards. Productivity tools proliferated. Knowledge worker productivity growth slowed.

And now, the fifth wave. AI.

ChatGPT reached one hundred million users in sixty days — the fastest adoption of any technology in human history. For the first time, technology can augment cognitive work — not just automate mechanical or repetitive tasks.

What AI actually does

Six things.

Cognitive augmentation. AI as a thought partner.

Skill democratisation. Non-coders building dashboards. Non-designers creating professional visuals.

Speed compression. Research synthesis going from four hours to twenty minutes.

Autonomous agents. Multi-step tasks executed without supervision.

Knowledge synthesis. Institutional knowledge encoded and applied at scale.

Personalised learning. AI tutors adapting to your pace.

The hard evidence — four studies

One. The BCG study with Harvard. 800 consultants. AI-assisted consultants completed 12.5% more tasks, 25% faster, 40% higher quality.

Two. GitHub Copilot. Microsoft study. Programmers completed coding tasks 55% faster.

Three. Stanford and MIT studied a customer service centre with 5,000 agents. AI assistance increased productivity 14% on average — but 35% for the bottom quartile. The AI compressed the experience curve.

Four. The Girotra study. ChatGPT-4 vs. elite MBA students. Independent judges scored every idea blind. 35 of the top 40 ideas came from the AI. Seven times more likely to rank in the top 10%.

Where this is showing up in BFSI

Underwriting. Insurers using AI to triage applications. Cycle times dropping from days to minutes.

Customer service. Indian banks running AI-assisted call centres in regional languages.

Investment research. AI for first-pass equity analysis.

Compliance and fraud. Pattern recognition at scale.

The firms not doing this are falling behind, mostly invisibly. Competitor's underwriting team processing twice the volume with same headcount. By the time it shows up in your numbers, you are three years behind.

Where AI falls short

One. Hallucination. AI confidently produces incorrect content. Garbage out, faster.

Two. Diversity. AI ideas are higher quality on average but less novel.

Three. Context. AI knows everything about the world in general — almost nothing about your specific organisation.

The Era of Abundant Intelligence

Now we get to the centre. Girotra calls this the era of abundant intelligence. It reframes everything.

For most of human history, intelligence was scarce. It lived in expert humans. AI has made a particular form of intelligence — pattern-based, language-capable, knowledge-synthesising — abundant. You can now get, at near-zero marginal cost, analysis that would have required a team of consultants.

This changes the economics of knowledge work as fundamentally as electricity changed the economics of physical work.

The historical pattern is striking when you lay it out.

1820s. Steam engine. Mechanical power abundant. Factories that redesigned became 10–100× more productive.

1890s. Electricity. Forty years before productivity gains showed up. Because factories had to be redesigned.

1990s. The internet. Information abundant. Encyclopaedia Britannica is gone. Google won — because they understood that when information is abundant, value moves to curation.

2023 onwards. AI is making intelligence itself abundant. The Britannicas of this wave will bolt AI onto old workflows. The Googles will redesign their entire model.

Where economic value goes

When something moves from scarce to abundant, value does not disappear. It redistributes.

When intelligence becomes abundant, value moves to four things:

One. Asking the right questions. AI answers — it does not ask.

Two. Evaluating AI outputs. When to trust. When to push back.

Three. Integrating AI judgments with human values, ethics, relationships.

Four. Decisions in genuinely novel situations the AI has no training data for.

What remains scarce

Original judgment in novel situations. Contextual wisdom. Authentic trust and relationships. Ethical responsibility. Accountability for outcomes can never be delegated to a machine.

The manager's new agenda

Five rules from Girotra's research.

One. Identify the right tasks to automate. High-frequency, high-cognitive-load, information-processing.

Two. Ensure information quality. AI is an amplifier. Garbage at scale.

Three. Redesign the whole process, not the task.

Four. Build human judgment back in at the right points.

Five. Make processes resilient. AI fails in novel ways.

The final thought

One sentence. Productivity has always been the central challenge of management. But the tools and the terrain have shifted more in the last three years than in the previous thirty.

The managers who thrive will be those who understand that their job is no longer to produce. It is to create the conditions under which other humans, augmented by abundant intelligence, produce their best work.

That is the real Compounding. And it starts on Monday.