State of Play: who is ready for AI agents in enterprise buying

We checked 648 companies on both sides of enterprise technology. One in nine startups lets an AI agent in. Not one large enterprise buyer does yet. The full database, sorted by agent-readiness.

The short version. AI agents are starting to research, shortlist and buy on behalf of enterprises. To take part, a company has to be readable by AI assistants and usable by AI agents. As of October 13, 2026, we checked 648 companies on both sides of enterprise technology: 449 startups and growth companies selling to enterprises, and 199 of the largest US enterprise buyers in the same sectors. The sellers are early. The buyers have barely started. The full database is below, searchable and sortable.

Who is ready

Cross-industry software went first. Notion, Linear, Vercel, Supabase, Rippling, Gong, Retool, Sentry, Calendly and WorkOS all run an MCP connector and publish an llms.txt. Their buyers already work inside AI assistants, so their agents arrived early.

Commerce is close behind. In retail and consumer goods, Algolia, Razorpay, Pine Labs, Chargebee, Gorgias, Amperity, MoEngage, Shiprocket and Postman are fully agent-ready, pulled along by payments and storefront partners.

Industry-specific technology is thin. In manufacturing, two companies out of a hundred run a connector: DroneDeploy and Katana. In healthcare it is six, led by Druva, MindTickle and Scrut, with Qure.ai, Luma Health and Turquoise Health also running one. In education, Brisk Teaching, Kiddom and Panorama Education. These are the sectors where buyers have the least time to evaluate vendors, and where an agent that can check a vendor’s claims would save the most.

The buyers

Not one of the 199 enterprises we checked runs an MCP connector on its main domain. One in ten publishes an llms.txt, among them Target, CVS Health, AutoZone, Campbell’s, Intel, GE Aerospace, Rockwell Automation, Cleveland Clinic, Cedars-Sinai, Penn Medicine, UCLA Health and Michigan State University. They are the early movers on the buying side.

That gap is the opportunity. When buyer agents and seller agents start meeting, someone trusted by both sides has to vouch for who is real, who fits and who wants to meet. That is the layer we are building.

The database

Every company we checked, sorted by agent-readiness. Search by name, filter by type or sector, or tap a column to sort.

How we checked

Two public signals, for each company’s primary domain. First, whether the site serves an llms.txt, the plain-text summary written for AI assistants. Second, whether the company runs a remote Model Context Protocol (MCP) server: we sent a standard MCP initialize request to the usual endpoints (mcp.<domain>/mcp and <domain>/mcp) and checked /.well-known/mcp. We counted a connector only when the endpoint answered like an MCP server, directly or with the MCP sign-in handshake. We did not sign in to anything. A connector on a different domain, or one offered only inside a client app, would not be counted.

What comes next

This is the first State of Play. We will rerun the check every quarter and publish what changed. If we missed your connector or your llms.txt, tell us where to find it at [email protected] and we will correct the record within a week.

If you sell to enterprises and you are not yet agent-ready, start with an llms.txt. It takes an hour. Our guide, Get found by the AI your buyers already use, covers it and the six steps after.

Method: automated public check of each company’s primary domain on October 13, 2026. Startups as ranked in SalesBook (methodology v0.2.7); enterprise buyers are 199 large US companies in retail and consumer goods, manufacturing, healthcare and education. Agent-ready score: MCP connector 60, llms.txt 40. Company-level public signals only.