GTM Company Brain

One line

An infinite canvas for GTM engineering, reimagined with generative UI inside Claude Cowork & Karpathy Wiki.

The problem

We run our entire go-to-market through AI — discovery calls, client research, proposals, outreach campaigns, content, positioning.
All through Claude. But there’s no operating surface and no shared memory.

We audited 60+ real Claude sessions across 8 months of our own GTM work and measured:

• 5-10 hours lost to re-explaining the same context,
• 70% of outputs throwaway because they died with the chat,
• and zero feedback loop between campaigns and strategy.

Founders entire GTM operation lived in scattered conversations that didn’t talk to each other.

Today

Seven motions repeat across every project. We identified them by reverse-engineering our actual work:

A call happens → we paste the transcript → Claude extracts stakeholders, pain points, decisions
We research a domain → market maps, competitor landscapes, workflow reconstructions
We turn positioning into outreach → campaign settings, enrichment prompts, email fragments
We prep for a meeting → company briefs, talking points, “never say” lists
We build a deliverable → proposals, decks, prototypes
We figure out positioning → founder thinking out loud, testing framings
We write content in our voice → LinkedIn, cold emails, blog posts

Every one of these starts from zero. Every one re-explains context that should be persistent.
The output of motion 1 should be the input of motion 3 — but it isn’t, because it died with the chat.

Tomorrow: Your GTM runs from one living canvas inside Claude.

A call happens → paste transcript → the wiki already knows the client, the stakeholders, the decisions.
The canvas shows what changed.
Click the deal that needs a proposal → type “build the deck” → first draft is 80% right because nothing was forgotten.
Select 6 prospects → “Create campaign” → fragments generate with your voice rules loaded → approve/reject in a review queue.
Results come back → the Retro Dashboard proposes updates → one click and the next campaign starts smarter.

Three layers make this work:

###Memory: Git-backed Karpathy wiki.
Raw sources in, LLM-compiled knowledge out. Companies, people, decisions, concepts — persistent, addressable, compounding.

###Interface:
Seven interactive nodes on a fullscreen MCP App canvas inside Claude. Deals, hypotheses, prospects, campaigns, results — spatial, clickable, live. Select in the canvas, direct in the chat, see the update.

Execution (roadmap):

Typed workflow runs sharing the wiki as common memory. Transcript processing, enrichment, campaign generation, retro analysis — each with a run ID, observable progress, wiki commits on approval. The hackathon proves memory + interface. Execution is what this becomes.

None, from scratch, of figuring out the concept.