AI Will Make IT Cheaper. I Believe It Will Make the World Grow Faster, Too.

By Karmendra Pandey

I spend my working life inside IT budgets, so let me start with what I can see from my desk, not from a stage.

AI is cutting the cost of IT work — not in theory, in measured studies. Customer support agents with a good AI assistant resolved 13.8% more issues per hour, and the least experienced agents gained the most, about 35%. Professionals in a controlled writing trial finished 40% faster with 18% better quality. Developers in large field experiments completed about 26% more tasks. And underneath all of it, the raw input is collapsing in price: getting a fixed level of AI performance cost $20 per million tokens in late 2022, and $0.07 about eighteen months later. When the price of anything falls more than 280 times, entire categories of work that were too expensive to automate — documenting every legacy system, summarising every incident, testing every line of old code — quietly become default.

That is the IT cost story, and I believe it. But it is not the story I actually want to tell.

The bigger claim

I foresee world GDP growing at a faster pace in the AI era than in the two decades before it. I know how that sounds, because the forecasters disagree wildly. The IMF projects AI adding roughly half a percentage point a year to global growth over the next decade. The OECD sees 0.4 to 1.3 points of extra productivity growth a year in the most exposed advanced economies. Goldman Sachs estimates a 7% lift to the level of world GDP over about a decade. And the leading sceptic, Daron Acemoglu, calculates barely 1–1.6% in total for the US over ten years.

Look at the spread and you learn something: nobody serious is arguing about whether AI works at the level of a task. They are arguing about how fast companies reorganise around it, how many tasks it can profitably do, and whether it creates new work or only automates the old. Those are questions about management, not about models. That is why I land where I do. An extra half a point of growth a year — the middle of the range — sounds modest until you remember that advanced economies have been stuck near 1–1.5% productivity growth for twenty years. Half a point, compounded, reverses most of that slowdown. We do not need the heroic forecast to be right for this to be historic. We only need the boring one.

History is the honest referee

Every great technology turning point looked disappointing before it looked inevitable.

Steam came first. James Watt's engine was patented in the 1770s, yet steam's peak contribution to British productivity growth arrived around 1850–1870 — roughly a hundred years later. Electricity was worse for the impatient: power stations appeared in the 1880s, but by 1900 electric motors still drove under 5% of factory machinery. The productivity boom only came in the 1920s, about forty years on, and only after factories were redesigned — one big steam engine and a forest of belts replaced by a motor on every machine, on single-storey production lines built around the flow of work, not around the power source.

Computers repeated the pattern so faithfully that economists named it. "You can see the computer age everywhere but in the productivity statistics," Robert Solow wrote in 1987. The statistics finally moved in 1995–2004, once companies redesigned processes around the technology instead of paving over old ones. The internet and the smartphone then taught the opposite lesson: adoption can be nearly instant — smartphones went from 5% to 40% of Americans in about four years — and the productivity boom can still not come, because consumer delight is not the same thing as reorganised production.

That history is exactly why I am not troubled that measured productivity has barely moved yet. AI is following the classic pattern, which researchers call the productivity J-curve: firms first spend on the invisible complements — data, workflow redesign, training, governance — which drag measured output down before the gains show up. Electricity needed new factories. AI needs new workflows. The technology is the easy purchase. The redesign is the work.

What is actually different this time

But I do not think this turning point is just electricity with better marketing. Four things are genuinely different.

First, speed. ChatGPT reached 100 million users in about two months. Electricity took decades to reach half of factories. AI rides on devices, networks, and cloud platforms that already exist, so there is no wiring of the world to wait for.

Second, copying is nearly free. A steam engine had to be built, shipped, and installed, one at a time. An AI capability, once created, can be copied to a million companies overnight at close to zero marginal cost. Intelligence has never before behaved like software. Now it does.

Third, it automates thinking, not muscle. Steam and electricity multiplied what hands could do. AI multiplies what minds can do — drafting, summarising, coding, analysing, supporting customers — including in the slow, expensive service sectors that resisted every previous wave of automation.

Fourth, it can help invent its own successors. AI systems have already predicted the structures of 200 million proteins, work recognised with a Nobel Prize. A technology that accelerates research itself is not just a tool in the economy. It is a lever on the rate at which the economy gets new tools.

The honest warnings

I hold this view, but I hold it with two warnings attached, and I have seen both from inside real IT organisations.

The first is that task gains leak. In one careful trial, experienced developers working in their own mature codebases were 19% slower with AI tools, while believing they were 20% faster. AI is fastest on bounded, pattern-rich work and can be a tax on deep, context-heavy work. Across a whole economy so far, the average worker's time saved works out to only about 1.4–2.2% of all hours — because most hours are not yet AI-assisted at all. Anyone who multiplies a laboratory gain by the whole of GDP is doing marketing, not analysis.

The second warning is about what happens to IT itself, my own industry. AI cuts the cost per unit of IT work far more reliably than it cuts the total IT budget, because demand expands to eat the savings — firms themselves estimate about 29% of cloud spend is wasted, AI creates new waste even as it removes old, and data centres are heading toward roughly doubling their electricity use by 2030. And for IT services firms like the one I work in, the business model itself is the thing being repriced: work sold by the human hour is work priced in the input AI cheapens. The future of my industry is outcome-priced, redesigned workflows, governance, and modernisation — selling the complements, not the hours. I consider that a healthy correction, but nobody should pretend it is painless, especially for the junior roles the old pyramid was built on.

My prediction

So here is my point of view, stated as plainly as I can state it.

Within this decade, world growth will run faster than it did in the sluggish decade before AI — not because the models are magic, but because the cost of intelligence is collapsing, the technology copies for free, and the redesign work, though slow, has already started in support desks, software teams, and back offices everywhere. The countries and companies that win will not be the ones that bought AI first. Electricity rewarded the factories that redesigned around it, roughly forty years after the power station. AI will reward the organisations that redesign around it — and it will not take forty years this time.

The steam age belonged to those who built engines. The electric age belonged to those who rebuilt factories. This age will belong to those who rebuild how work itself is organised, around a mind that can be copied. That is a turning point at least as large as the previous ones — and for the first time in history, it is arriving faster than our institutions are ready for. That gap, not the technology, is the real risk. And closing it is the real opportunity.

References

1. Brynjolfsson, E., Li, D. & Raymond, L. — "Generative AI at Work," NBER Working Paper 31161 / Quarterly Journal of Economics (2025): support agents +13.8% issues/hour, +35% for least-skilled. https://www.nber.org/digest/20236/measuring-productivity-impact-generative-ai
2. Noy, S. & Zhang, W. — "Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence," Science (2023): writing time −40%, quality +18%.
3. Cui, K. et al. — field experiments with AI coding assistants across 4,867 developers (2024): +26.1% completed tasks.
4. Peng, S. et al. — "The Impact of AI on Developer Productivity" (2023): 55.8% faster on a standardised coding task. https://arxiv.org/html/2302.06590v1
5. METR (2025) — RCT with experienced open-source developers: 19% slower with early-2025 AI tools in their own repositories.
6. Stanford Institute for Human-Centered AI — AI Index Report 2025: inference cost at GPT-3.5-level performance fell from $20 to $0.07 per million tokens (Nov 2022–Oct 2024).
7. IMF (Oct 2026) — Georgieva: AI could add ~0.5 percentage points to annual global growth over the next decade. https://www.imf.org/en/blogs/articles/2026/10/15/the-world-cant-afford-to-get-ai-wrong
8. OECD (June 2025) — "Macroeconomic Productivity Gains from Artificial Intelligence in G7 Economies": +0.4–1.3 pp/yr labour productivity in high-exposure G7 economies. https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/06/macroeconomic-productivity-gains-from-artificial-intelligence-in-g7-economies_dcf91c3e/a5319ab5-en.pdf
9. Goldman Sachs (2023) — "Generative AI Could Raise Global GDP by 7%." https://www.goldmansachs.com/insights/articles/generative-ai-could-raise-global-gdp-by-7-percent.html
10. Acemoglu, D. (2024/25) — "The Simple Macroeconomics of AI": US GDP +1.1–1.6% in total over 10 years. https://economics.mit.edu/news/daron-acemoglu-what-do-we-know-about-economics-ai
11. McKinsey Global Institute (2023) — "The Economic Potential of Generative AI": $2.6–4.4 trillion/year across 63 use cases.
12. Bick, A., Blandin, A. & Deming, D. — Federal Reserve Bank of St. Louis: AI users save 5.4% of their own hours, ~1.4–2.2% of all work hours.
13. Crafts, N. — steam's contribution to British productivity growth peaked c.1850–70, ~100 years after Watt; ICT's contribution exceeded steam's.
14. David, P. (1990) — "The Dynamo and the Computer": electrification's ~40-year lag and the factory redesign it required.
15. Brynjolfsson, E., Rock, D. & Syverson, C. — "The Productivity J-Curve," American Economic Journal: Macroeconomics (2021).
16. Flexera — State of the Cloud 2026: organisations self-estimate 29% of cloud spend wasted.
17. IEA — "Energy and AI" (2025): data-centre electricity 415 TWh (2024) rising to ~945 TWh by 2030.