Who gains from AI, who pays for it, and what India should do about the difference.
This page grew out of two talks delivered at the Takshashila Institution in June 2026. Both talks are preserved below exactly as delivered; this guide is the version that will keep growing over time. Feedback is always welcome, please feel free to leave a comment below.
AI is a general-purpose technology, so the invention is not the story — the reorganization is. Steam, electricity, and the internet changed society by changing the production map, through complements built around them over decades.
It is expensive to build and cheapening fast to use — and most bad AI arguments come from picking only one of these facts. Frontier training costs have compounded like a capital project while token prices collapsed by orders of magnitude. Both are true at once.
The AI stack is a chain of required returns. Chip and power suppliers are already earning; compute operators must out-earn depreciation; token buyers must find ROI; the economy must show diffusion. Each layer's revenue is the next layer's cost — validation lives at the top.
Depreciation is where the boom is actually decided. Whether a GPU usefully lives two years or five is the most boring number in AI, and it settles whether today's buildout turns a profit or becomes a write-off.
Value created is not value captured. The ledger only sees what gets billed; consumer surplus leaks out of every layer. AI can enrich the world and still disappoint the people who financed it — ask the railway shareholders.
AI changes tasks first; jobs, firms, and statistics move later. Productivity shows up when contracts and org charts change — Coase and Cheung, not benchmarks, explain why the macro numbers lag the demos.
There is no job apocalypse in aggregate — there is concentrated stress at the entry rung. Early-career workers in exposed occupations show a measurable relative employment decline, while solo founders scale like never before. Both are the same reorganization, seen from different rungs.
India is exposed where it earns. The IT-services arbitrage — many people per dollar of revenue — sits in AI's early blast radius, and the first shock will be local, urban, and occupational, not national and abstract.
India is vulnerable, but has optionality. English, an IT base, digital public infrastructure, and a huge domestic market are real assets — if institutions make good experiments cheaper and turn them into ladders. Vulnerable is not doomed.
The future is a coordination problem, not a forecast. Measure early, experiment fast, insure humanely — and read every AI headline by asking which ledger changed, and what evidence would prove it.
The guide
The six parts, and where each one lives.
This is the map. The ten ideas above are the argument; these six are the doors into it. Each part is being written up from the talks; until it moves in, its links take you to exactly where it lives in the talk — and Part 1's exhibit is already embedded below.
1
What kind of technology this is
The machine
A general purpose technology is actually the easy bit. Rearranging our civilization around it is the hard part. To better understand AI, you have to understand other general purpose technologies from the past: electricity and the internet are great examples. What those GPTs did to work, and the bargaining around work helps us understand what is likely to happen this time around... but with one crucial difference. This GPT's capability frontier moves very, very quickly.
The exhibit below is live: an interactive time-lapse of model capability against price. Drag the timeline; the frontier is the most capability you could buy at each date. Open the chart full-screen →
2
Value created vs value captured
The money
Capex is expensive in AI. But the good news is that inference is becoming cheaper by the day. The real thing to keep track of when people talk about the AI stack is whether each stack's revenue generates enough for the whole thing to keep humming. And the thing that really decides this is... depreciation.
AI changes tasks before it changes jobs, and contracts before it changes statistics. The evidence so far: broad usage, concentrated stress at the entry rung, and firms changing shape.
How will India be affected? Depends on which India you are talking about. The bad news is that India is exposed most right where it shines: IT and ITeS. The good news, on the other hand, is that India also has real options. But as always, there's an asterisk, and that's where the reading matters.
Measure the shock before it is obvious, make good experiments cheaper, keep frontier access open, and treat the safety net as a state-capacity problem. Easier said than done? You bet!
There's economics, which is confusing but exciting. Then there is accounting, which is boring, but clarifying. The bad news is that you have to keep both stories on track in your head.
Parts appear here as they are written; the talks below remain the complete argument in the meantime. The library and the open questions are already live.
The talks
As delivered at Takshashila, June 2026.
The two seminars this guide grew out of, preserved exactly as presented — dual-mode pages that read as essays and present as slides.
Here's a curated, reasonably-frequently-updated (no promises!) reading list, related to this topic. Have I read every word in each of these? Hell no. Should you? Also hell no. Exercise judgment, and happily have your LLM of choice read the longer ones. But curate your reading, and never ask the LLM to choose for you — that'd be my way of going about it.
The empirical engine room: value migrating from the infra layer to the model labs, who now charge by the economic value of the token. Maps directly onto the four-layer ledger. Paywalled past the midpoint.
The counterweight that keeps the guide honest: a falling labour share is not a transfer from labour to capital, and real compensation is at an all-time high. The worry is the future Jevons world, not the present tense.
The missing India vantage on compute: a consumer-integrator with supply access but little capability ownership. The best person I know working on AI and semiconductors in India.
Acemoglu & Johnson · Annual Review of Economics 16, 2024
The canonical this-has-happened-before brick: British real wages roughly halved 1806–1820 and shared prosperity took decades plus political voice. The productivity bandwagon is a choice, not an automatic.
Travel-agent employment fell over 60% from its peak, yet surviving agents' relative pay rose from 87% to 99% of the private-sector average. The machine took the separable part of the bundle and left people the strong part.
The demand-side complement to Garicano: Starbucks automated for years, concluded it was a mistake, and is re-hiring baristas for handwritten notes on cups.
“The relational sector grows precisely because it isn't automated.”
The optimist's structural-change rebuttal to humans-as-horses — my counter: the induction argument is just saying the sun will rise tomorrow because it always has. One day it won't. A ready-made dialectic.
Managers make good workers in an AI-first world: agentic tools amplify returns to judgment and allocation rather than flattening them. Verify the linked PDF before quoting specifics.
The freshest as-we-speak evidence of the shift to delegated production: 5× weekly users in H1 2026, ~10× more 8h+ delegations since January. Caveat: OpenAI-internal usage is the frontier, not the average firm.
Philip Trammell · Substack (with Dwarkesh Patel), Dec 2025
The cleanest mechanism for who captures value: while labour is a bottleneck (Baumol) its share can rise; once it isn't (Jevons), the capital share climbs toward one. A post, not a working paper.
“Piketty was wrong about the past. He's probably right about the future.”
The most uncomfortable, most India-relevant implication in the literature: when capital substitutes for labour, the engine of catch-up growth shuts off. Say it plainly.
Kim (INSEAD) & Koning (HBS) · HBS WP 26-090 · SSRN, June 2026
Microdata for how firms reorganise: AI-native startups are 25% smaller, 13% more engineers, ~15% fewer entry-level workers and managers — comparable valuations, more value per employee. Cited in both talks; one treatment in the guide.
Tedeschi, Rama & Cruickshank · Stripe Economics, June 2026
The n=1 limit of the AI-native firm and the optimistic flip side of the canary: ~4m Americans earning $100k+ solo. The India question — solo-scaling on DPI rails — should be stated as a question, not a finding.
Alter Magazine · alter (drawing on Bhattacharya, Deaton, Rajagopalan)
Before asking what AI does to Indian jobs, ask whether we can measure Indian jobs. Bonus vignette: when the US blocked computer sales, the ISI built India's first computer in 1953 from war-surplus parts.
“A country that could not buy computing technology built its own from the debris of wars.”
A century of human-capital divergence: China bottom-up, India top-down. Education inequality is ~25% of wage inequality in India vs under 12% in China — why India did services and China did factories.
The crisp backbone for engineers-vs-lawyers: talent that organises production innovates; talent that rent-seeks redistributes. Lands hard with an Indian professional audience.
The engineering-state vs lawyerly-society thesis, heavily highlighted in my copy. Pair with Hessler's Other Rivers and Ang's How China Escaped the Poverty Trap.
Rodrik & Stiglitz · The New Global Economic Order (Routledge, 2025)
The missing macro spine for the India half: manufacturing became an enclave, services must absorb labour but hit a home-market ceiling — and AI now threatens the services escape hatch too.
The single most on-thesis India-AI-labour case: ~$50k revenue per employee vs a US benchmark near $300k. India's services-arbitrage model is the thing AI most directly erodes. (Original is paywalled; the headcount economics are corroborated in Lightspeed's open write-up.)
Ground-level texture for India climbing the manufacturing ladder: a dormitory-labour regime traced back to 1920s Shanghai cotton mills. The vivid counterweight to abstract labour-share macro.
Nandlal Mishra & Pramit Bhattacharya · Data For India
The rigour footnote for any manufacturing-employment claim: household surveys say 68.5M by 2022–23, enterprise surveys a static ~48M. Keep every claim defensible.
Telangana's Composite Backwardness Index across 242 caste groups: the strongest predictor of backwardness is access to English-medium education, not land. The bridge from caste to who reaches the AI economy.
Jesús Fernández-Villaverde · X thread (with John Burn-Murdoch)
TFR already 1.88, projected to US levels by ~2031, Kerala 1.3 vs Bihar 2.9. The dividend window is closing faster and unevenly. Not a crisis yet, but it is coming.
First causal national estimate: no positive spillover to US-born workers from immigration enforcement. The India parallel is sons-of-the-soil politics meeting internal migration.
The policy-menu source: token taxes, sovereign wealth funds, automation adjustment assistance, the physical-vs-human-capital tax bias. My caveat: the scenario buckets are not well defined.
Economic Survey 2016-17, Ch 9 · Government of India
The UBI argument is really a state-capacity argument: a reliable cash floor plus portable benefits plus active placement, delivered by a state that can actually do those things.
The Hayekian reading of Amara: the short-run shortfall is mundane friction; the long-run surprise comes from dispersed experiments no planner could design. Exactly the PSST point, made rigorous.
“Anybody who says they predicted this is selling something.”
The named optimist to argue with, not a strawman: finite compute means opportunity cost governs deployment. He concedes the three cracks — inequality, adjustment frictions, AI capturing its own means of production.
The honest caveat before prescribing India's next bargain: history only happens once, so there is no science of development. Humility on which lever is decisive.
Above water, not ashore: trailing revenue ~$110bn, ~19% depreciation headroom on stated assumptions. The best single synthesis to date-stamp and revisit.
Ethan Mollick · X (on the BBC 'Industrial Revelations' series)
Mollick's one-line version of this guide's opening claim: steam changed nothing until thousands of skilled workers worked out how to apply it to their existing jobs. Swap steam for AI and that is the adaptation problem exactly — the technology is the easy part; the reorganization is the long grind. He points to the BBC 'Industrial Revelations' series as the illustration.
“Steam alone wasn't enough, it required thousands of skilled workers figuring out how to apply this new power to their old work.”
The machineThe work
Nothing matches — clear the search or pick another part.
Open questions
The questions I left with the room.
A field guide that only asserts is not worth rereading. These are the questions the talks ended on — verbatim — and the ones this guide is still working on. The question is not "Will AI help India?" The question is: which Indians, which cities, which firms, and which ladders?
01
If AI is better at many tasks, what should humans become relatively best at?
from The Economics of AI
02
If AI can do 40% of your tasks, which 60% becomes more valuable, and which 20% should never have been in your job?
from AI, Work, and India
03
If a hospital, bank, or startup can be unplugged from a frontier model overnight, what is that model worth on its balance sheet?
from The Economics of AI
04
If only approved institutions can touch frontier capability, who gets to decide what AI is for in Indian conditions?
from AI, Work, and India
05
Design education around projects where students must use AI, explain what they trusted, show what they rejected, and defend the final judgment. What would that look like in an Indian classroom?
from AI, Work, and India
06
When you hear an AI headline, which ledger changed — and what evidence would prove it?
the question to leave with
About this guide
A living document, not a post.
For ten years I wrote blogposts about economics. This is what I now believe replaces most of them: one growing, navigable document instead of a scattered archive — with the talks that seeded it preserved exactly as delivered, and every claim carrying its source.
The guide grows the way my reading works: I clip an article, argue with it properly, and then it earns a place here — as a library entry, as evidence inside a section, or occasionally as a new section. Updates follow my reading rhythm, not a publishing calendar; the changelog above is the honest record of both.
The register throughout: steel-man the optimists, keep the sharpest dissent, and hold two true things at once. Where the evidence is genuinely unsettled, the guide says so instead of picking a team.
Written by Ashish Kulkarni, built with Claude. Sources marked ✓ were checked against primary documents during talk preparation (June 2026). Single self-contained page; the only external call is to Google Fonts.
Changelog
What changed, and when.
A living document should show its work. Every meaningful addition lands here first, and the best of them become posts on the blog.
3 July 2026
econofai begins. The hub: a ten-idea TL;DR, the six-part spine with the falling-price-of-intelligence chart embedded live, both June 2026 Takshashila talks preserved as delivered, a 55-source reading list with an outbound link on every entry, and six questions for the room.
Comments
Leave a comment.
Feedback, corrections, and pointers to what I've missed all make this guide better. A threaded comment section is on the way; until it lands, the fastest line is email — ashish@econforeverybody.com.