# Tim Hey

> Field notes from inside a real agentic GTM function. Agent discovery, Agent Experience, and the shift from selling to humans to selling to machines.

Tim Hey, Sr. Director of Product, Agent Discovery at Zapier.

Every page here is available as raw Markdown — append `.md` to any URL, or send `Accept: text/markdown`.

## Writing

- [The eval that tells you if agents actually pick you](https://timhey.co/posts/the-ask-the-model-eval?via=llms): You can ship every artifact and still lose. The only way to know if agents choose you is to ask them, repeatedly, and score it.
- [One registry will not get your MCP server found](https://timhey.co/posts/get-your-mcp-server-found?via=llms): Agents find MCP servers two ways: in the registries their client reads, and on the open web mid-task. Most servers do the first and miss the second.
- [An agent cannot call an API it has to guess](https://timhey.co/posts/an-api-an-agent-can-call?via=llms): Give an agent a machine-readable contract or it guesses, gets 400s, and falls back to a rival with a clean spec. The spec is your API's README.
- [An SDK an agent can find, install, and run](https://timhey.co/posts/the-agent-ready-sdk?via=llms): An SDK is found as a package and judged by its docs. Distribution and docs are one chain, and the agent quits at the first broken link.
- [The package registry is the agent's search index](https://timhey.co/posts/the-registry-is-the-search-index?via=llms): An agent looks for a tool by querying a registry, not Google. No package under a name it can predict, no candidate. The decision is made before your content gets a vote.
- [Most businesses don't need most of the agent files](https://timhey.co/posts/which-agent-files-you-need?via=llms): The list of files agents look for is long, and most of it is not for you. A capability file is only worth serving if you have the capability behind it. How to tell, and what they look like.
- [The files an agent checks before it reads your page](https://timhey.co/posts/the-files-agents-look-for?via=llms): Agents probe a fixed set of well-known paths before they read a word. Serve them and you hand the agent your rules, your map, and your tools.
- [The industry standardized agent discovery. A spec is not adoption.](https://timhey.co/posts/a-standard-is-not-adoption?via=llms): Eleven companies, Google and Microsoft included, agreed on how agents find your tools. On launch day almost none of them had shipped a catalog. Here is what actually changed.
- [An agent reads your page as text. Most of it is gone.](https://timhey.co/posts/what-an-agent-reads?via=llms): Agents do not render your JavaScript or see your layout. They read the page as a string. Here is how to make sure the answer survives.
- [Getting found by an agent is a distribution problem](https://timhey.co/posts/agent-discoverability-playbook?via=llms): Agents do not browse, they query. Here is the six-channel playbook for being the tool an agent picks, and how to measure whether it works.
- [Your GitHub repo already logs agent interest. Read it.](https://timhey.co/posts/github-agent-traffic?via=llms): GitHub won't label which visitors were agents. But clones, referrers, and paths give you the signal. A four-read how-to off the traffic API.
- [Agentic GTM is two jobs wearing one name](https://timhey.co/posts/two-disciplines?via=llms): GTM run by agents and selling to agents are different disciplines. Most people conflate them. The split is the whole point.
- [The trust gap nobody is closing](https://timhey.co/posts/the-trust-gap?via=llms): Everyone optimizes the site they own. The data says agents weight independent corroboration. That is the unclaimed move in agentic GTM.

## More

- [Resume](https://timhey.co/resume?via=llms): career history, roles, and competencies.
- [What agents read](https://timhey.co/agents?via=llms): a human-readable index of every machine-readable file this site serves.
- [ai-catalog.json](https://timhey.co/.well-known/ai-catalog.json?via=llms): ARD capability catalog for this site.