Part IRun GTM as a product
A harness is your interface with AI
Ask four companies what a harness is and you get four answers. One is Claude chat with the company's connectors switched on. Engineers work in Cursor, with rules in the repo. Ramp gives every employee a workspace already set up, with skills colleagues share. Owner.com's revenue team built the whole stack and keeps it in GitHub.
GTM teams need a shared harness for AI.
Like a laptop, every new hire should get it on day one.
In the Field Guide: What is a shared AI harness?
Design AI GTM around your buyer, not your 2020 org chart
Making each role faster leaves the handoffs between them in place. A shared harness lets more of the team answer the buyer directly.
With specialization, you can hire the perfect person. But now you have these handoffs that are very inefficient for the customer.
…the buyer gets expert depth at the first touch instead of four handoffs later.
The point is fewer handoffs for the buyer, not fewer job titles.
A harness starts as a repo and works as a loop
A simple harness is a repo on GitHub: a CLAUDE.md that says how the team works, and a folder of skills.
$ git diff skills/sales-deck/SKILL.md
- Start with our company overview.
+ Start with the buyer's problem.The harness isn't a repo. It's your GTM product loop.
In the Field Guide: Fix the harness, not the output
AI builders run the harness as a product
The harness grows past a folder of skills, and somebody has to build and run it.
An MCP tells Claude what a system can do. A skill tells it how this team uses that system. The harness needs both halves.
They're the product group for the harness.
Sam Gong, who founded Fresh Context, shows one we built. Studio, our brand image service, broke on 1 October when the vendor's image model stopped making transparent backgrounds. One fix in Studio's data, and every request since comes back clean, for everyone.
In the Field Guide: Builders own it as a product. Everyone else runs it
The boxes are squishy. Anyone can build now
Building isn't only for engineers now. A seller mocks up the screen they need in Lovable in an afternoon. A marketer asks Claude for a skill that updates a CRM field.
Then, live and with no engineer, Sam asks Claude for a skill that logs his Sculpt meetings to HubSpot, runs it on one real meeting, and the record shows up.
Just like engineering, GTM needs product leadership and space to rebuild around AI.
That's GTM product thinking: someone owns the bottleneck, and anyone in the trenches can help fix it.
In the Field Guide: What is GTM product thinking?
The map is the product
One last layer, systems of action, completes the map.
Every box is a place to ask: what's the bottleneck, and who fixes it?
Three GTM workflows on one harness
Sam runs three GTM workflows live, on the same harness.
- Event invites from anywhereEach teammate invites their own network, tracked like a BDR program.
- Studio and brandAnyone on the team makes and swaps images from one brand system.
- ABM pages updated with new diagramsOne skill updates every live account page with the new diagram, then confirms it's done.
The map, the harness and the builders, run as a product.
Your first skill
Pick the task someone on your team did by hand twice this week. That's your first skill.
Name who fixes it when it comes out wrong, and count the days. The principle is in the Field Guide entry, What is a shared AI harness?
Draft copy, agent's words, not yet reviewed.