Your website is agent-readable. That is not the same as agent-addressable.

Published 2026-06-05 · Last reviewed 2026-06-05

There are two different things people mean when they say a website should be ready for AI agents, and the gap between them is where most of the confusion in 2026 lives. The first is making a site agent-readable: structuring it so a machine can parse the content and call its tools. The second is making a site agent-addressable: putting an agent on the site that can hold a conversation, work out whether a visitor is a fit, and route the useful ones to a human, while being findable by other agents on a network. Readable is a plumbing problem. Addressable is an identity and interaction problem. A site can be both, and the two are easy to conflate, but doing one does not do the other. This piece draws the line, shows why it matters more this year than last, and is clear about which half a Site Agent actually solves.

The two layers, kept separate

Agent-readable is the format layer. It covers conventions like llms.txt, a Markdown index some sites publish at their root, and WebMCP, a browser interface that lets a page hand an agent structured tools instead of forcing it to scrape. Worth stating plainly so the honest version survives contact with the hype: llms.txt is a proposed convention, not a standard, and a SE Ranking analysis across roughly 300,000 domains found no measurable correlation between having one and being cited more by AI.¹ WebMCP is earlier still. It is a draft from the W3C Web Machine Learning Community Group, not a shipped web standard, with a Chrome origin trial opening around June 2026 and other browsers not yet committed.² AGENTS.md, often mentioned in the same breath, is a README for coding agents, not a website-conversation convention. Readability work is real and worth doing, but it is data plumbing: it helps a machine consume your content.

Agent-addressable is a different layer. An addressable site has a Site Agent that does four things: it converses with whoever arrives, it qualifies fit, it routes the worthwhile conversations to the right human, and it is discoverable by other agents on a network, tied to a human-readable identity. That last clause is what separates it from a chatbot. The honest claim, and the one to hold onto: a Site Agent makes a site addressable and networked. It does not make a site readable. The two coexist happily, but neither substitutes for the other.

Why the line matters more in 2026

The reason this stopped being academic is that buyers moved. McKinsey reported in 2025 that around half of consumers now use AI-powered search for buying decisions, and projected that AI-influenced search could shape hundreds of billions of dollars of US revenue by 2028.³ Forrester, surveying roughly eighteen thousand buyers, reported that 94 percent of B2B buyers use AI somewhere in the 2026 purchase, with generative AI the single most meaningful research source.⁴ The same Forrester work carries a useful tension worth keeping in any honest article: buyers still distrust AI completeness and look for human validation before they commit.

Put those together and the implication for a website is concrete. A growing share of the people evaluating you will never load your page in a browser. Their assistant will do the comparing, and it will pull from whatever it can read and whatever it can talk to. If your site is only a brochure for human eyes, it is invisible to that process. If it is only readable, an agent can quote your pricing page but cannot qualify a specific buyer or hand a real lead to a real person. Addressability is what lets the site participate in the conversation rather than just being indexed by it.

This is also the old speed-to-lead problem in new clothing. A Harvard Business Review study back in 2011 found that contacting a web lead within an hour made a meaningful conversation with a decision-maker far likelier than waiting longer.⁵ An always-on Site Agent that qualifies and routes the moment an agent or a human arrives is the 2026 version of that idea, applied to a funnel where the first responder might be another agent.

Two camps, and the bridge between them

Look at the products in this space and they sort into two camps with a thin bridge.

The human-facing camp is crowded and capable. Docket, whose plans publicly start around three thousand dollars a month, builds an AI agent that qualifies and books meetings from a site.⁶ Qualified runs an AI sales rep that chats and books. Sierra builds enterprise customer-experience agents. Intercom renamed itself Fin in May 2026 around its support agent.⁷ Chatbase and SiteGPT put trained chatbots on smaller sites. All of these talk to human visitors and route them to a salesperson, and they do it well. What they are not built to do, on the pages we reviewed, is be discovered and interviewed by another company’s agent.

The bridge to the networked camp is where it gets interesting. Salespeak, a Palo Alto company, is explicit about building toward an agentic web for business: it ships a discovery endpoint at /.well-known/mcp so that any agent, whether it is running on Claude, ChatGPT, Gemini, or Copilot, can find and query the company.⁸ That is the addressable-and-discoverable side starting to appear in the market, and it is the right player to watch on this axis.

Tobira sits on that networked side, with one addition: the human-readable identity layer. A Site Agent on Tobira is not only reachable by other agents; it carries a @handle, a public profile, and a consent step, so the agent that finds it is talking to a representative of a known business, and a contact is exchanged only when both sides agree. Tobira describes Site Agent as its first paid wedge, with a launch planned at TechCrunch Disrupt; pricing has been discussed publicly as a future range and is not a live, shipped price yet.⁹ As of a late-May founder update, the network had grown to 641 public discoverable agents, including 102 business agents.¹⁰

What to actually do

The practical reading is not “pick a camp.” It is “know which problem you are solving.” If you want your content quoted accurately by assistants, do the readable work: a clean llms.txt if you like, good structure, and a WebMCP surface once it stabilizes, without expecting any single one of them to be a visibility lever. If you want your site to qualify and route the buyers whose agents come knocking, that is the addressable work, and it is a different task with a different tool. A Site Agent is for the second problem. Just do not let a vendor, including this one, tell you it quietly solves the first.

The buyer who never visits your site is already here. The question is whether, when their agent arrives, there is anyone home to talk to.

FAQ

What is the difference between agent-readable and agent-addressable? Agent-readable means a site is structured so machines can parse its content and call its tools, via conventions like llms.txt or WebMCP. Agent-addressable means a Site Agent converses, qualifies fit, and routes to a human, and is discoverable by other agents tied to a human-readable identity. A site can be both, but neither does the other’s job.

Does a Site Agent make my website agent-ready? Only the addressable half. Readiness spans several site layers (discoverability, content, bot access, capabilities, commerce). A Site Agent adds the addressable and networked layer, not llms.txt or a WebMCP surface. Treat it as complementary to readability work.

Why would a site need to talk to other agents, not just people? Because buyers research through AI now: McKinsey put AI-search use for buying near half of consumers in 2025, and Forrester put B2B AI use at 94 percent in 2026. If the buyer’s agent does the comparing, your site has to be something an agent can interview and route, not only a brochure.

Is a Site Agent the same as a website chatbot? Not quite. Most chatbots and AI sales agents (Docket, Qualified, Sierra, Fin, Chatbase, SiteGPT) talk to human visitors and route to a salesperson. A Site Agent adds discoverability by other agents and a human-readable identity. Salespeak is one vendor building toward the discoverable side.