The questions people ask first.

Tacit is an AI studio that writes a company's unwritten knowledge, the rules its best people apply without thinking, into skills its AI agents follow. We work in four stages: Audit, Foundry, Review and Readable. You own everything we produce.

A skill is a folder an AI agent reads before it does a task: a SKILL.md file with the rules, the steps and the checks, plus any reference files and scripts it needs. It is plain text, readable by a person and followable by agents such as Claude and Codex, and it lives in your own repository. Ours ship with an eval set, so you can prove the skill still works when the model changes.

A business is agent-readable when an AI agent acting for a customer can find what it offers, check its rules and book it without guessing. That takes structured, current facts on your own website and a booking path an agent can complete. It is to AI assistants what search engine optimisation was to search engines.

An automation agency builds workflows that encode the task. We write down the judgement behind the task, so the agent handles the cases a workflow cannot, and we test it against your own examples. Automations get cheaper every month as agents learn to build them; knowing what to tell the agent does not.

That is the model, and it is good. What it does not have is any idea how your company works: how you quote, what you refuse, who gets called first. Our work is writing that down as skills the agent follows, with tests that prove it still does. Keep the tool you have; we make it yours.

No. The skills are your people's judgement, written down, and the review policy keeps the decisions that matter with them. What changes is how much of the repetitive part they do by hand. The person who held the knowledge is the author of the skill, and remains the one who can change it.

Every skill ships with an eval set: real cases and the output your expert accepted. When a new model arrives, you re-run the set and see in ten minutes what still holds and what broke. The skills themselves are plain Markdown, not tied to any one model or vendor. Monthly care does the re-run for you if you would rather not.

For the audit, no. We work from real examples you send us: emails, quotes, tickets. For Foundry we need access to the agent product you use, in your account, so the skills are installed where they will run. For Readable we need your website's repository or hosting, or we hand you the files.

The eval score and the outputs side by side. During the build you see every output the agent produced, including the wrong ones, and you watch the score move as the rules are corrected. We do not quote efficiency percentages or hours saved. You can count those yourself from the digest.

Each stage is scoped and priced after the first call, as a fixed amount for a fixed scope, not an hourly rate. The audit is the smallest and comes first so you can judge the work before committing to more. Prices are not published here because they depend on how many tasks and departments are in scope.

Audit: two weeks. Foundry: three to five weeks per department. Review: one to two weeks. Readable: two weeks. The stages run in order, and you can stop after any one and keep everything produced so far.

Any where the valuable knowledge is held by people rather than written down, which is most of them. The method is the same everywhere: find the tasks, interview the experts, write the rules, test them. The rules are yours and specific; the method is not.

No. Oakfield Plumbing is a fictional scenario, built so the page can show what a skill, an audit row and an offer file look like without exposing anyone's real rules. The sentence you type into the home page stays in your browser and is not sent anywhere.

We do not build a new agent product. We make the one you use know your business. If you do not have one, we recommend one for your size and say why, and we install into it.

Yes. Many teams do the audit and stop, with a task map and a constraints file they use on their own. Foundry needs an audit first, because the rules come from it. Review and Readable can each be done on their own if you already have skills or an offer to describe.

Remotely, in your working hours. Interviews are video calls, everything we produce is text in a repository you control, and the handover happens with your people at the keyboard. A site visit is not needed for any stage.