The Action Economy: A Thesis for Building Foundational AI in India
Vijaya Mahadevan is in her early 50s, a digital marketing educator with a small Tamil Instagram following, and a habit of explaining technology through the life it actually touches. When she explains AI agents, she throws in an example of building a software that tracks the weather via an API, notifies cloud movements for rain, and sends a message so that you could run upstairs and remove the dried pappad from the terrace.
While the west always teases the possibility of booking your dream Europe trip via AI agents, in India all you might need is a notification to save your pappad.
The AI for the next billion people probably doesn’t require a better model. It requires a clearer understanding of what a common person actually needs to do.
How AI Evolved - and Why India Needs a Different Entry Point
Global AI products had a clear progression
First came the chatbots - conversational interfaces that answered questions.
Then came AI agents - systems that could take multi- step actions on your behalf.
Then came connectors - agents that plugged into your existing tools, calendars, CRMs, email, and orchestrated across them.
And now, we’re watching the emergence of the all- in- one general- purpose assistant like OpenClaw and Hermes - a single agent that can handle anything you throw at it.
In the west, this progression makes sense. Information was already organised. Workflows already lived in digital systems. AI could plug into a structured world and make it faster.
India needs to skip all of that.
We are jumping from unorganised to AI - without fully going through the intermediate step of organising data and information through digital systems.
This is the structural reality that most AI labs in India will face: you can’t build a connector when there’s nothing to connect to.
India Solves Problems Through People, Not Processes
In India, you get advice for free. You pay for the action. Free advice from the mutual fund distributor - you pay for the transaction. Free recommendation from the pharmacist - you pay for the medicine. Free guidance from the gym owner - you pay for the membership.
The value chain in India has always terminated in action, not information. And the action has always required one of two things: a person who does it for you, or you becoming the person who figures it out yourself.
What This Means for Architecture
Cost is the first argument people reach for. An Indian user pays ₹500 a month; a frontier model burns that in a day. True. But cost only tells you what you can’t afford - it doesn’t tell you what to build. The real case has nothing to do with price.
A frontier model is built for breadth. It can write a poem, debug Python, and explain thermodynamics, because nobody knows in advance what you’ll ask it. That breadth is the product. It’s also the cost.
But the business owner every month isn’t going to ask it anything unpredictable. He has one task. File the GST. Send the reminder. Reconcile the ledger. The job isn’t “answer anything.” It’s “do this one thing right, the same way, ten thousand times.” That’s not a breadth problem. It’s a reliability problem.
The inversion that changes everything is this: The standard fear is that a small Indian model can’t compete with the giants - they’re bigger, they’ve read more, they have a head start. But read what? The public internet, in English, mostly Western. The kirana workflow was never written down. The thing you actually need the model to know was never digitised - so the frontier model’s head start is worth exactly nothing here. It never saw the data either.
The Action Economy doesn’t need a smaller brain. It needs a model that knows the one thing the big brains never learned.
Two Pains, One Pattern
Two things are happening simultaneously in India right now. Both are accelerating, and both point to the same structural gap.
Pain at Work: The People You Need Are Getting Harder to Find
A woman running a tailoring business in Nagpur has the same operational needs as a 50-person company - marketing, accounting, customer management, inventory, operations. She can’t afford a single person for any of it.
Tier 2 agencies will promise all of it for a thousand rupees a month and deliver none of it well. The economics of competent help simply don’t work at her scale.
This isn’t a niche problem. India has an estimated 63 million MSMEs. The vast majority are single- operator or family - operated businesses where one person plays five roles - badly. They’re not underserved by technology. They’re underserved by people. And the gap between what they need and what they can afford is widening every year.
Pain at Home: The Systems You Need to Navigate Are Getting Harder to Figure Out
Seekho has over 100 million downloads. It’s a paid app. Its most popular courses teach people how to update Aadhaar, renew a driving license, file for a birth certificate. People are paying real money to learn how to do things that should be simple.
An Indian household doesn’t just consume. It administers. Government documents, school admissions, insurance claims, property paperwork, tax filings. Every one of these systems was designed for institutional capacity - not for an individual person navigating it alone on a phone.
The pattern across both pains: the task requires either a person you can’t afford, or knowledge you can’t act on. Both end in the same place - done badly, or not done at all.
The Buildability Test: A Framework for Prioritisation
Not every use case is worth building. Before you spend engineering time, run each candidate through four filters:
Painful - they’re doing the task badly or skipping it entirely. They’re bleeding.
Recurring - daily or weekly, not once a year. Frequency is what pays back the learning curve. A birth certificate is painful but happens once; social media happens every day.
Visible - the outcome is a concrete artifact. A post published, an invoice sent, a form filed. No artifact, no proof it helped, no way to charge.
Monetisable - the user is economically active and the value lands in rupees. Time saved, revenue recovered, fines avoided. Measurable economic value is what a product can price against.
The best AI products for India won’t try to create new behaviour. They’ll replace broken behaviour - the tasks people already do badly, often, with visible outcomes and real money at stake.
What We’re Actually Building
Go back to the pappad on the terrace for a second.
The woman in that story is lucky, though the story never says so. Someone in her life knew how to turn a problem into a few lines of code. Most people don’t have that someone. That’s the whole essay, really - it’s about everyone standing on the terrace with the rain coming in, and no one to call.
Because in India, getting things done has always meant knowing the right person. The one who knows which counter to go to. The one who’s filled the form before. The one you slip a little extra to outside the government office because the system was never built for you to face it alone. If you have that person, things happen. If you don’t - they happen badly, or they don’t happen at all.
Hand each of them something that just does the thing - reliably, in their language, for the price of a recharge. This is the one thing the giants can’t build for us, because they don’t have our problem. Their models learned a world that already wrote everything down. Ours hasn’t been written down yet. So the field is empty - right now, today. Empty fields don’t stay empty.
Build for them.
Dharmesh, very interesting perspective. One thing wasn't clear for me though: Should India build "models" using the unorganized data we have? Or should we pick up open-weight models and repurpose them with our organized data, on top of which we build apps like the papad example? As I type this, I also wonder - in either of the cases - how do we convert the unorganized data into something useful?
This also reminds me of ChatGPT's campaign during this year's IPL season. It was built around every day tasks of common people: fixing a water filter, readying documents while applying for a bank loan etc.. all built around the tasks we feel we need guidance for. The marketing focuses on simple adoption across existing needs. Hits the right note to reach many.