The Open API Revolution. Survival of the Crawlable.

I have been selling enterprise software for 20 years. SaaS, CPaaS, MarTech, AdTech — into some of the largest organisations in India. Banks. Telcos. Government agencies. Fintechs. In that time, I have watched three technology shifts change who wins and who becomes irrelevant.

The first was the internet. The question it asked of every company was: do you have a website? Companies that said no did not just miss a channel. They became invisible to the infrastructure that was routing customers, capital, and attention. The web could only find what was indexable. Everything else stopped existing commercially.

The second was open source. Linux, MySQL, Apache, Android. The companies that built on open infrastructure won distribution. The ones that tried to wall off their technology found that the community gave the same thing away for free. Economic value migrated up the stack — to whoever could best serve the layer above the commoditised one.

We are now in the third shift. And the question it is asking is one most Indian technology companies have not thought about yet.

"Is your product crawlable? Can an AI agent find you, connect to you, and act through you? Because if it cannot — agents will choose your competitor. Not because your product is worse. Because your competitor is reachable."

This is the Open API Revolution. And unlike the previous two shifts — which played out over a decade — this one is moving in months.

The Context Window Is the New Search Index

Here is the mechanism. Understand this and the rest of the essay clicks into place.

When a human searches Google, Google presents ten options. The human reads, evaluates, clicks. The decision is human. The action is human.

When an AI agent operates on your behalf — booking a hotel, placing a supply order, resolving a support ticket, executing a GST filing — the agent does not search and show you options. The agent decides. It calls the services available to it in its context window, and it acts.

The context window is the entire world the agent knows about. If your product is not in that context window, the agent does not consider it. Not because it evaluated and rejected you. Because you did not exist.

This is not hypothetical. Look at what happened in the last 30 days.

Stripe launched Link in April 2026 — a digital wallet built explicitly so that autonomous AI agents can shop, pay for reservations, buy tickets, and execute transactions on behalf of users. With OAuth authentication, spend controls, real-time approval flows. Stripe did not build this for developers. It built it because it understood that the primary consumer of payment infrastructure is increasingly going to be an AI agent, not a human typing in a card number.

Swiggy launched Builders Club at mcp.swiggy.com/builders. Three MCP servers. 18+ API tools across Food, Instamart, and Dineout. Open to developers and enterprises. Swiggy is not doing this for developer goodwill. It is doing this because the next generation of food orders will be placed by AI assistants — "order my usual from Swiggy" — not by humans opening apps. If an AI assistant cannot connect to Swiggy's MCP server, it cannot place that order. A competitor that is connectable will capture that order instead.

Swiggy even hosts llms.txt and llms-full.txt at the domain root. This is the equivalent of robots.txt — but for AI agents. It tells every AI crawler: here is what we are, here is what you can use, here is how to connect. It is a small detail that signals a very deliberate choice.

The GSMA Signal at Scale

The GSMA Open Gateway initiative is the most visible enterprise example of this realisation at scale. As of 2025, 73 operator groups representing 285 networks and 80% of global mobile subscribers have committed to a common API framework — standardised access to network capabilities like SIM swap, number verification, fraud signals, and quality-on-demand. McKinsey estimates that if operators successfully expose these APIs, they unlock an additional $300 billion market opportunity by 2030.

They are not being altruistic. They are making their infrastructure crawlable before someone else makes them commodity pipe. The operators that move fastest on fraud signals, quality-on-demand, and mobile payment APIs will capture that $300 billion. The ones that wait will find that hyperscalers have built the integration layer above them and they have become infrastructure everyone runs on without paying for.

The protocol that makes all of this possible — MCP, the Model Context Protocol — has gone from 1,200 public servers in Q1 2025 to 9,400+ in April 2026. In December 2025, Anthropic donated it to the Agentic AI Foundation under the Linux Foundation umbrella. The same move Google made with Android. You cannot use it as a moat now. Every company that builds an MCP server is building on permanent, open infrastructure. Which means the moat is not the protocol. The moat is being there first.

Open Source Was About Code. Open API Is About Access.

The structural parallel matters because it tells you where the value migrates.

Open source commoditised a layer of software — operating systems, databases, web servers. Value moved up the stack, to whoever could build the best service on top of the commoditised layer. Red Hat did not sell Linux. It sold enterprise reliability on top of Linux. Amazon ran open source better than anyone and charged for the convenience. Google gave Android away and monetised everything above it.

The Open API Revolution commoditises a different layer. Not the software itself. Access to the software. The companies that expose clean APIs, publish MCP servers, and make their data and functionality composable for AI agents are becoming the infrastructure layer that the new ecosystem runs on.

The value, again, migrates to whoever best serves the layer above.

"The companies that win the Open API era will not be the ones with the best products. They will be the ones whose products are in the most agent workflows."

The Crawlability Stack — Where Are You?

Not all API exposure is equal. I have been using a simple framework in conversations with clients. I call it the Crawlability Stack. Four layers. Know which one you are on.

Layer 1 — Discoverable. You have API documentation. Developers who specifically look for you can find you. This was adequate in 2020. An AI agent will not read your documentation. It will query a registry, check for an MCP endpoint, and either connect or move on in milliseconds. If you are at Layer 1, you are already behind.

Layer 2 — Connectable. You have an MCP server. Standardised authentication. Any AI agent using the protocol can call you. This is the 2026 baseline. Swiggy is here. Stripe is here. Atlassian, Figma, Notion, HubSpot — they are all building here. If you are a SaaS company without an MCP server on your roadmap, you have a product gap that will become a commercial gap within 18 months.

Layer 3 — Composable. You do not just expose your own capabilities. You become the orchestration layer through which AI agents compose multi-step workflows. Stripe's Link is a Layer 3 play — it is not just a payment API, it is the trusted financial interface through which agents manage entire commerce workflows across dozens of merchants. GSMA Open Gateway's 43 channel partner companies are Layer 3 plays. Value concentrates here.

Layer 4 — Training Data. You have proprietary data that AI models want to train on. This is the long game. A bank's credit history. An operator's fraud signal library. A hospital's clinical outcomes at scale. The most underexploited category in the GSMA Open Gateway is currently mobile payments — 1% of deployments despite strong developer demand. The operator that builds it first has a dataset nobody can replicate.

The New Pricing Models

When access becomes the product, pricing changes. Per-seat pricing — which Blog 3 argued is structurally dying — gets replaced by something more granular and more honest about where value actually sits.

Inference fees. You pay per call. Every time an AI agent queries your API, a micro-transaction occurs. This is already the model for every major LLM. It will extend to vertical data providers. A compliance API providing real-time GSTN status checks will charge per query. A fraud-signal API will charge per verification. Unit economics shift from monthly seats to per-inference revenue.

Agentic revenue share. When an AI agent completes a transaction on a user's behalf, the platform that facilitated the access gets a cut. Stripe's Link is built for exactly this model. The commerce platforms that win this decade will not be those with the best consumer apps. They will be those whose APIs are running inside the most agent workflows.

Training data licensing. Proprietary data gets licensed to model builders. This is already happening in media — News Corp, Axel Springer. It will extend to telecom fraud signals, financial transaction patterns, clinical outcomes. Operators who have treated their data as a compliance burden are about to discover it is the most valuable asset they own.

Crawl credits. A company with proprietary data grants AI agents the right to index and use it in exchange for payment or reciprocal data. llms.txt is the first visible form of access control that will eventually become a pricing layer. Swiggy already has it. Most Indian companies do not know it exists.

India: The Most Interesting Paradox

India occupies a position that no other market does. It has some of the best public API infrastructure in the world, deployed at national scale. And simultaneously, one of the weakest API-first cultures in its enterprise software market. The gap between those two facts is the largest untapped opportunity in Indian technology.

On the infrastructure side: UPI processes 18 billion transactions a month. Aadhaar has authenticated 1.4 billion identities. GSTN processes filings for 14 million businesses. ONDC is building open commerce rails that no platform can capture. These are not legacy systems. They are Layer 3 composable infrastructure at national scale. Any AI agent operating in India that needs to verify identity, process a payment, or check tax compliance will have to call these APIs.

"India Stack is the best public API infrastructure in the world. Indian enterprise software has almost no MCP-ready products. The company that builds the MCP layer on top of UPI, GSTN, and Aadhaar — making India's compliance infrastructure composable for AI agents globally — will be worth more than most current Indian SaaS companies combined."

On the enterprise side: the picture is almost the opposite. Most Indian SaaS companies treat their API as a feature for developer integrations, not as the primary product. Most CPaaS companies have APIs designed for humans to read documentation and write code — not for AI agents to discover and call at inference speed. Swiggy is one of the very few exceptions.

There is also a specific risk for Indian enterprise buyers. If the global SaaS tools they use — Salesforce, SAP, Workday — build MCP servers that connect to global AI agent ecosystems, those agents will operate from a context window that does not include Indian compliance data. The agents will be built on Western data, trained on Western workflows, and optimised for Western business contexts.

A procurement agent that can execute a purchase order through a global SaaS MCP server but cannot verify GST compliance through GSTN before releasing payment is not a productivity tool. It is a liability. Indian CIOs who do not demand India Stack connectivity in their AI vendor evaluations are setting themselves up for a compliance incident they will not see coming.

What You Should Actually Do Before the End of 2026

The window for being an early mover is closing. Here is the prescription, by role.

If you are a SaaS founder or product leader:

Does your product have an MCP server? If not, this is a P0 priority, not a future consideration. 30% of enterprise app vendors will ship MCP servers by end of 2026. If you are not in that 30%, you are in the 70% making a bet that your category is immune.

Have you implemented llms.txt? Takes hours. Signals to every AI crawler what you are and what you allow. Swiggy has done it. Have you?

What is your inference fee model? If you are still thinking exclusively in annual contracts, you do not have a pricing strategy for the agent era. Build a per-call pricing tier before your competitors do.

If you are in enterprise sales (SaaS or CPaaS):

Stop selling integrations. Start selling crawlability. Every enterprise buyer you talk to is already using AI agents or will be within 18 months. The question they should be asking — and you should be answering before they ask — is: can your product be called by our AI agents?

Map your product to the Crawlability Stack before your next sales conversation. Layer 1, 2, 3, or 4. Know where you are. Know where your competitors are. The delta is your sales narrative.

For BFSI, government, and regulated verticals: demand MCP connectivity to Indian compliance APIs as a procurement requirement. Any AI tool that cannot connect to GSTN, UPI, or Aadhaar is a tool that creates compliance exposure the moment agents start executing regulated workflows.

If you are a CIO or IT leader:

Audit your vendor stack for MCP readiness. Every SaaS tool you are paying for should be asked: when is your MCP server shipping? Make it a renewal clause, not a polite question.

Build or commission an internal MCP server for your enterprise data. Your internal knowledge base, compliance data, and operational metrics need to be in your AI agents' context windows. If they are not, your agents will make decisions without your most important context.

Evaluate India Stack connectivity in every AI tool evaluation. An agent that cannot natively connect to UPI and GSTN is not built for Indian enterprise. It is a Western tool with an Indian price tag.

The Companies Reading This Correctly

Stripe: repositioned itself as the financial infrastructure of the agentic internet. Link is not a product feature. It is a bet on which company gets to sit in every agent's context window when a payment needs to happen.

Swiggy: one of the only Indian consumer platforms to have built an explicit MCP layer. Three servers, 18+ tools, co-branding for builders, direct engineering support. The message is clear: we want to be the infrastructure that food-ordering agents run on, not a destination humans open on their phones.

GSMA Open Gateway: 73 operator groups, 285 networks, 80% of global mobile subscribers. The operators that move fastest on fraud signals, quality-on-demand, and mobile payment APIs will capture $300 billion. The ones that wait will find that hyperscalers have built the integration layer above them.

Anthropic: by donating MCP to the Linux Foundation, they made the same move Google made with Android. You cannot use it as a moat now. Every company that builds an MCP server is building on permanent infrastructure.

Before You Close This Tab

The Open Source Revolution did not destroy software companies. It forced them to move up the stack — from selling code to selling integration, from licensing software to delivering outcomes. The companies that moved won larger markets. The ones that defended their existing model became infrastructure everyone ran on without paying for.

The Open API Revolution will do the same. It will not destroy software companies. It will force them to move from building products humans use to building infrastructure AI agents run on. The companies that make that move will win markets that do not yet exist. The ones that defend their current model will find that agents have simply routed around them.

"Open source was about making code free. Open API is about making access universal. In both cases, the winners were not the ones who gave things away. They were the ones who understood what the next layer of value looked like — and got there first."

Swiggy understood this. Stripe understood this. GSMA understood this. Anthropic understood this. The question for every Indian technology company reading this is not whether the Open API Revolution is coming. It is already here.

The question is which side of it you will be on.

---

Being easy to integrate is the new competitive advantage.

Being crawlable is the new distribution.

Being in the context window is the new market share.

---

The Intelligence Economy — A Four-Part Series: Blog 1: One Person. One Model. One Company. Blog 2: The GPU Wars. Why Silicon Is the New Oil. Blog 3: The SaaSpocalypse. $2 Trillion Gone. Blog 4: The Open API Revolution. Survival of the Crawlable. ← You are here