Claude for GTM: how GTM teams run on Claude

Claude for GTM means running your go-to-market from one Claude brain at the centre of the stack. It knows your company, your competitors and your metrics. It reads and writes to every tool you sell with, and it improves with every campaign. People keep the strategy, the approvals and the closing.

This is how we run every client account, and how we run our own agency. Below is what the brain holds and the layers it runs on. Then what Claude does well, where a person stays in charge, and how the setup grows with the team.

01What running GTM on Claude means

For years our stack was a row of tabs. Sourcing lived in one, enrichment in another, copy in a third and sending in a fourth. The real work was carrying data between them by hand.

That changed when Claude Code moved into the middle. It reads and writes to each tool directly, so the whole system runs as one loop. At the centre of that loop sits what we call the Claude brain. It analyses, creates and helps the team create.

Every client gets their own brain, in a separate GitHub repository. On day one it holds what we gathered at the start. Depending on the client, that means sales calls, CRM data, website visitors, and content and ad metrics.

Then every campaign, message and reply goes back into it. By month three, the brain holds the things that usually live in people's heads:

• The ICP as it converts, set against the ICP we started with.
• The copy that worked and the copy that died, each version next to what it produced.
• The offer framing this particular buyer responded to.
• Every reply and its tag: interested, not now, wrong person or not interested.
• The client's own language, taken from what their buyers say on sales calls.
• Each change to targeting or messaging, with the reason written down beside it.

When a person leaves an account, that context leaves with them. A chat session forgets it by Monday. The brain keeps it, because it lives in files that stay with the account. You can see what goes into one on our GTM brain page.

02The four layers of Claude for GTM

A brain needs infrastructure under it. Ten leads and ten thousand leads run on the same reasoning. What changes with volume is whether the data stays current, the loops fire on time and the credentials hold.

Knowledge sits in GitHub. Every client has its own repository, and their whole GTM lives in files. Because it is versioned like code, nothing important depends on one person's memory. Routines pull in what happened across Slack and ClickUp each day, so Claude reasons from this week.

The brain is Claude Code. It runs the sourcing logic, the scoring, the copy, the campaign builds and the reply handling. Before any job, it reads its instructions and the client's context from the knowledge layer.

Loops run on Trigger.dev and n8n. Trigger.dev decides what fires and when. n8n handles the glue between tools. Clay still runs enrichment inside outbound, and Claude orchestrates across all of it.

Memory is Supabase. It holds the lead and account data, and every other part of the system checks against it. Each cycle starts from what the last one learned.

Underneath all four, the keys live in a vault. A resolver makes each call, and Claude only ever receives the result.

Put together, a signal from RB2B or Trigify becomes a sourced lead in Prospeo. BounceBan verifies the email, Smartlead queues the message and Attio records the pipeline. Supabase remembers each state, and Claude carries the record from one stage to the next.

This is the same system behind $18M+ in client revenue across the campaigns we run.

03Claude for GTM engineering: who builds it

Someone has to design these layers and keep them running. That person is the GTM engineer, and the discipline is GTM engineering. Claude for GTM engineering means building with Claude Code, versioning the work in GitHub and testing it like software.

On every account we pair the engineer with a GTM strategist. The strategist chooses the segment, the offer and the signals worth acting on. The engineer turns those choices into loops that run every day. Claude speeds up both jobs and replaces neither.

04What Claude does well, and where a person stays

Claude is strongest at work that repeats and needs a lot of context. A mistake that cannot be undone is where it needs a person. We draw that line in the same place on every account.

Every system we build holds that line. A skill drafts, reports or stages the work, and then it stops. A person approves before anything reaches a prospect or spends money.

Three rules hold the gate in place. Claude never sees a password. It cannot send an email on its own. Purchases carry a budget cap, and larger ones need a second confirmation.

The grading logic sits in code, so the same numbers always give the same verdict. Claude reads that verdict and explains it. It never re-tunes the model halfway through a run.

05Skills: how the method reaches every account

The brain holds what a client knows. Skills hold how we work. Each skill is a single markdown file that teaches Claude one job, from list building to reply triage. It names the tools it may open and the knowledge to read first. It also marks where it hands over to a person.

Our methodology runs as one skill tree across seven phases, from setup through daily operations. When we fix a skill once, every client gets the fix the same afternoon. The client's facts never live in the skill, so nothing crosses from one account to another.

We publish 87 of these in our Claude skills directory. For a shortlist, start with the best Claude skills for GTM, grouped by the job each one does.

06How the setup changes with team size

The setup that works for a solo founder breaks at five people. The five-person setup breaks at twenty. What changes is the plan Claude works from, the stack it reaches and how much runs unprompted.

We run the twenty-person version internally. Two operators briefing the same model from memory end up building two different systems. That is why the plan moves into files at five.

07The five levels of AI in GTM

A useful way to place a team is by what AI does in production this week. We see five levels between a team that uses AI and a GTM system that runs on it.

1. 01AI is a tab. Someone opens Claude, pastes something in and closes it. The process looks the same afterwards.
2. 02AI is a feature. Apollo recommends contacts and HubSpot summarises the deal. Nothing talks to anything else.
3. 03AI is a layer. Clay enriches the record and Claude drafts the copy. Every output is still a human decision.
4. 04AI is an engine. A job change triggers enrichment, copy and a queued send, chained together by n8n.
5. 05AI is the brain. Claude decides what runs next, and Supabase holds the memory between cycles.

Running Claude for GTM is level five. It starts with using Claude to reason through the strategy itself, with drafting as only one of its jobs.

08Where to start with Claude for GTM

1. 01Write down what the company knowsThe ICP, the offers, the proof, the objections and the CRM history go into one repository before anything is automated.
2. 02Run one job as a skillPick a job that repeats every week, such as account research or call prep, and turn it into a skill file.
3. 03Add memory and one loopConnect Supabase so results are kept. Then schedule the one loop that costs the most time by hand.
4. 04Put the gate in before you scaleCredentials go in a vault, sends sit behind a human sign-off and spend sits behind a cap. Then add the next layer.

Call prep and prospecting are good first jobs. Our beginner's playbook covers the hands-on side: which Claude surface to open, which plan to buy and how to connect your tools.