Two-key GTM

What Two-key GTM is

Two-key GTM is a way to run go-to-market work with AI agents in which nothing ships on one key. An agent drafts or acts, which is the first key. A second key, another agent or a person, checks the work against its evidence before it reaches a customer, a budget or a production system. I named it after the two-person rule used where one mistake is expensive: two separate keys, held by two separate people, both needed to act.

The three rules

Evidence attached. Every claim carries its source, and a claim that can’t be checked doesn’t ship. In practice the first key hands over the work and its evidence together: the citations behind a blog post, the screenshot behind a competitor brief, the query behind a number.

Agents draft. People decide. Agents handle research, QA and routing. A person approves what goes out to customers, what spends money, and what changes a production system. The second key belongs to whoever answers for the outcome.

Systems over heroics. A campaign that works should run twice, so it becomes a skill, a checklist or a pipeline with gates. A one-off win that nobody can repeat isn’t a system yet.

How the keys work

The first key is a drafting or acting agent with a written acceptance criterion. It produces a result and the evidence for it in one handoff. In my stack that agent is usually Codex or Claude Code running a named skill.

The second key is a different reviewer. For a draft, it’s a quality gate with a score threshold and a rewrite brief routed back to the step that owns the fix. For a task on the shared agent hub, it’s the other model or me. For a change to a production system, it’s me, and my approval has to be less than 30 minutes old.

Two-key GTM keeps one writer per production system: Google Ads, HubSpot, Customer.io and n8n. Only one agent may change each system at a time, and every change goes into a journal. Any other agent reads the hub and sends its change to that writer instead of touching the system.

What it costs

The second key costs time, and review becomes the constraint once agents produce at speed. My next change is to route the second key by risk: agents review agents for reversible internal work, and a person reviews only what ships, spends, sends or changes a production system.