The Agent Economy: What Happens When Software Has a Wallet

Software has always been a tool. You tell it what to do, it does it, you review the output. The human is in the loop at every decision point.

That's changing. And the change isn't about better AI — it's about giving AI economic agency.

The Wallet Primitive

A wallet is the most underrated primitive in the agent stack.

Without a wallet, an AI agent is a sophisticated recommendation engine. It can research, compare, negotiate — but it can't act. It finds the best flight but can't book it. It negotiates a vendor rate but can't pay. It identifies a tax-saving investment but can't execute.

With a wallet, the agent becomes an economic actor. It moves from advisory to executive. This is a category change, not a feature addition.

The Three Economies

Economy 1: Human-to-Agent (H2A)

This is where most people start thinking about agent payments. A human delegates tasks to an agent, the agent spends money on the human's behalf.

Examples in India today:

Personal finance agent that automatically invests in SIPs via BSE Star
Shopping agent that buys household essentials when inventory runs low
Travel agent that books the cheapest Rajdhani ticket when prices drop below a threshold

The economic model is straightforward: the human sets a budget, the agent optimizes within it. The value proposition is time savings and better execution.

Market size: every Indian consumer with a smartphone and a willingness to delegate. Start with the 100 million Indians who already use UPI daily.

Economy 2: Agent-to-Agent (A2A)

This is where things get genuinely novel. Agents hiring other agents, paying for services, forming temporary economic relationships.

Consider a content creation workflow:

Your marketing agent determines you need a blog post about tax changes
It discovers a research agent that specializes in Indian tax law (via an agent marketplace)
They negotiate: ₹2,000 for a 1,500-word research brief, delivered in 2 hours
Research agent delivers. Marketing agent verifies quality against criteria
Payment releases from escrow
Marketing agent passes the brief to a writing agent
Writing agent produces the final post

Seven steps. Zero human involvement in the execution. The human set the initial goal — "keep our blog updated on tax changes" — and the agent economy did the rest.

The economic implications are profound:

Friction-free specialization. Agents can be hyper-specialized because coordination costs approach zero
24/7 labour markets. Agent workers don't sleep, don't take holidays, don't negotiate benefits
Instant scaling. Need 100 research agents for a time-sensitive project? Spin them up in seconds
Granular pricing. Services can be priced per task, per minute, per output — no annual contracts

Economy 3: Agent-to-Infrastructure (A2I)

The most overlooked category. Agents that autonomously procure and manage infrastructure:

Server scaling: agent monitors traffic, spins up cloud instances in Mumbai region, negotiates spot pricing
Data procurement: agent identifies useful datasets on India's Open Data platform, purchases premium extensions
API consumption: agent manages quotas across map APIs, translation APIs, payment APIs — optimizing for cost and performance

This economy is largely invisible to humans but critical to the agent ecosystem's efficiency.

The Pricing Revolution

Human services are priced on proxies. We charge hourly rates, annual salaries, project fees — because measuring actual output is hard.

Agent services will be priced on outcomes:

Human Pricing	Agent Pricing
₹5L/year for an accountant	₹50 per tax return filed
₹2,000/hour for a lawyer	₹500 per contract reviewed
15% fee for a recruiter	₹10,000 per qualified candidate
₹1L/month for a marketing team	₹100 per lead generated

This isn't a prediction — it's an inevitability. When the service provider is software, input-based pricing makes no sense. You pay for what you get, not for the time it takes.

The implications for India's service economy — which employs hundreds of millions — are worth taking seriously. More on this in a future essay.

The ₹100 Crore Question: Transaction Volume

Let's model what agent transaction volume could look like in India by 2027.

Conservative assumptions:

10,000 businesses using agent payment infrastructure
Each business has 5 agents making payments
Each agent makes 20 transactions per day
Average transaction: ₹2,000

Daily: 10,000 x 5 x 20 = 1,000,000 transactions Daily GMV: ₹200 crore Monthly GMV: ₹6,000 crore Annual GMV: ₹72,000 crore

These are conservative numbers. They assume only business agents, not consumer agents. They assume modest transaction counts. They assume no A2A economy.

With the A2A economy factored in — where agents are transacting with each other, creating multiplicative effects — the actual volume could be 5–10x higher.

The Infrastructure Requirements

Supporting an agent economy requires infrastructure that doesn't fully exist today:

Agent Identity. Every agent needs a verifiable identity — who created it, who authorized it, what are its capabilities. Think of it as PAN for software.

Agent Discovery. Agents need to find other agents. This requires registries, reputation systems, capability descriptions. India's ONDC framework — originally designed for e-commerce discovery — could be adapted for agent discovery.

Payment Rails. UPI provides the settlement layer. But agents need additional infrastructure for:

Escrow management
Multi-party payments
Conditional releases
Dispute resolution

Audit Infrastructure. Every agent transaction needs a complete audit trail. Not just the payment, but the reasoning behind it. This is critical for:

Tax compliance (TDS, GST)
Regulatory reporting
User trust
Dispute resolution

The Network Effects

Agent economies have powerful network effects:

Direct network effects. More agents on the network means more potential counterparties for any individual agent. Your marketing agent has more research agents to choose from. Your procurement agent has more suppliers to negotiate with.

Cross-side network effects. More buyer agents attract more seller agents, which attract more buyer agents. Classic marketplace dynamics.

Data network effects. More transactions generate more data about pricing, reliability, and quality. This data makes every agent on the network smarter. The platform with the most transaction data can offer the best agent-matching algorithms.

Ecosystem network effects. Developers build specialized agents for platforms with the most users. More specialized agents attract more users. This creates lock-in at the platform level.

The first platform to reach critical mass in India's agent economy will be extremely difficult to displace.

What This Means

India has three structural advantages in the agent economy:

Payment infrastructure — UPI provides instant, low-cost settlement at a scale no other system matches.
Developer talent — India produces the world's largest number of software developers. The talent pool for building agents and agent infrastructure is unmatched.
Market size — 1.4 billion people, rapidly digitizing. The addressable market for agent services is enormous and growing.

The question isn't whether India will have an agent economy. It's whether we build the infrastructure to capture the value — or let it be captured by platforms that don't understand our rails, our regulations, or our market.

This is the infrastructure question of the decade. And the window to answer it is open right now.