← Home
the school of future
01 / 14
use ← → keys
orientation session · SOF class of '28 · july 2026

empire of AI

Prantik Barua
co-founder & COO, the school of future

first, where this title comes from

a book that is critical of AI

Empire of AI, Karen Hao, 2025. A hard, critical look at the industry, told from inside OpenAI. I named this session after it on purpose: to study any technology seriously, start with the people questioning it, not the people selling it.

picking up from yesterday

two threads, one session

building for bharat · akshay

90% of India's workforce, close to half of GDP, runs on the informal economy. Street-level fluidity, no credit histories, business models that change every festival. An entire economy of unsolved problems.

alphago · last night

Move 37: the machine finds what no human would. Move 78: Lee Sedol answers with the divine move, because the machine pushed him there. The machine makes you better, or it makes you lazy. How you hold it decides.

today: the map of who builds AI, where the game is open for you, and the habits that decide which side of Move 78 you end up on.
the map

the AI value chain

the map · tap each layer

the AI value chain

chips & compute
cloud & infrastructure
models
tooling & orchestration
applications & deployment bharat's layer
↑ narrower means fewer players, more capital, deeper moats
why the bottom layer is not the kids' table

india: the use-case capital

At the application layer, the moat is not the model. It is local knowledge, distribution, and trust. UPI proved an Indian deployment layer can become world-defining infrastructure. Nandan Nilekani's argument: India should not chase frontier models, it should win deployment at population scale.

you hold ground truth about tier-2 and tier-3 India that nobody in San Francisco, and honestly nobody in Bengaluru, has. that is not a consolation prize. that is the moat.
connecting it back

every problem akshay named
is an empty square on that layer

1
Creditworthiness with no credit history. 500 million workers the banking system cannot score. An AI underwriting problem, waiting.
2
Business models that change every month. The street's fluidity needs tools that reconfigure as fast as a festival calendar. Corporates take 5-7 years; the street takes a week.
3
An equal-opportunity workforce. Voice-first, low-literacy-friendly tools for the segment that adopts everyone.
4
5,000 companies of 2-3 crores, not one or two unicorns. The empire of many. That is the real GDP engine, and the real opening for this room.
proof the game has begun

the scramble

Reliance
New subsidiary Reliance Intelligence (2025). Partnerships with Google Cloud and Meta, AI across telecom, retail, energy, finance.
Mahindra
Mahindra.ai, a dedicated group AI division, driving strategy across autos, farm, finance, hospitality.
Godrej
₹1,200+ crore committed over 3-5 years to digital and gen-AI across the enterprise group.
Adani
Two years ago: 49:51 JV with Sirius (Abu Dhabi's IHC), focused on AI, IoT and blockchain.
read it correctly: they are spending hundreds of crores to buy access to a capability that is sitting in your pocket tonight, for free. the gap is not the model. the gap is the operating habit.
the uncomfortable half

the machine produces the artifact.
it does not produce the judgment.

You get Claude accounts this week. I am not telling you not to use them. I am telling you every evaluation here tests judgment: specificity, framework, depth, honesty. AI-generated work survives until exactly one follow-up question from a mentor. Use it to sharpen what you did. The moment it replaces what you didn't do, the second question will find you.

the toolkit

habits, not prompt tricks

Prompt hacks go stale in a quarter. Operating habits compound for a career. Four habits, and one structure to hold them.

1
Bring something true. AI sharpens a real thought. It cannot manufacture one. Arrive with what you observed, decided, or built.
2
Make it argue with you. Ask for the strongest case against your plan before you ask for help polishing it.
3
Context is the multiplier. The quality of what you get is set by what you feed it. Thin brief, thin answer.
4
Close the loop yourself. Take its output into the field, test it against reality, bring back what actually happened. That last step is the part that makes you valuable.
your structure · four projects, four rooms

set up your claude like this

project 01

SOF Term 1

Modules, frameworks, vocab, session prep. Your course brain. Feed it what you learn each week.

project 02

Missions & Labs

Field work on real businesses. Keep the mess here: notes, drafts, what the shopkeeper actually said.

project 03

Hidden Brand

Your capstone. Separate on purpose, so its thinking compounds privately over 18 months.

project 04

Personal Projects

Your own writing, ideas, side builds. The room that is only yours.

why separate rooms: context is the multiplier, and mixed context is diluted context. each project gets smarter about one thing.
the window

for the first time, the kid in a small town holds the same tools as the kid three streets from the head office.

this has never been true of any previous technology. it will not stay this open for long.

so what is an empire of AI?

not one giant.
five thousand builders.

The empires on the news are buying capability. The empire that matters for Bharat is Akshay's number: thousands of 2-3 crore companies solving real problems on the application layer, built by people who know the ground. That is the empire this room can build.

they paid crores.
you hold it free.
build the habit.

go build