How does AI change go-to-market work?

A daily judgment round. Five short questions a day, one area at a time, answered with a tap by the people who own that judgment. It collects more judgment from more people than a review meeting does, and it takes each of them two minutes.

A weekly lint. Once a week, something reads the team's shared material and flags contradictions, stale claims and gaps, and a person decides what to fix. Andrej Karpathy's LLM wiki pattern (April 2026) makes lint one of three basic operations, because upkeep that used to defeat every team wiki now costs almost nothing.

A brief before any draft. Who reads it, what they believe now, what they should believe after, and what they do next. The brief carries more of the quality than the draft does.

Examples in the prompt, prohibitions at the gate. We tested it on our own pages: a plain model given only a good brief beat every page we had launched, six times out of six. The fix was to move the list of things never to say out of the prompt and into a check that runs after the draft, and to give the model approved examples instead.

A default and a date on every decision. Each open call carries a recommendation, a default and the date the default lands. Silence becomes an answer, and work stops waiting on the busiest person in the room.

A named owner for each generated surface. Elena Verna, writing about working at an AI-native company (September 2026), calls this the agent parent: the person responsible for the quality of what an agent produces. Much of what used to be management, she argues, was the work of moving information through a hierarchy.

Reviewers who judge rather than write. The reviewer's job becomes deciding between drafts and teaching the system why. That's a different skill from writing, and it needs to be hired for.

An owner for the context itself. Someone keeps the team's shared knowledge current, the same way someone keeps the CRM clean.

A change that touches money, customers or the law gets checked before it lands, not linted after. Lint on a schedule is for the rest: the large, cheap, reversible body of work that used to wait in an approval queue.

A starting point for a team moving to this way of working:

• Name an owner for each AI surface the team already uses.
• Pick one body of shared material and lint it every week.
• Start a daily judgment round for one kind of output.
• Put a recommendation, a default and a date on every open decision.
• Write a brief before the next piece of work, and keep it beside the work.

The rituals scale better than the meetings they replace. A judgment round reaches a hundred people as easily as five, and a weekly lint covers a content library no approval committee could read. What doesn't scale on its own is ownership: every surface needs a named person, and in a large company that means writing the role into job descriptions, not hoping someone picks it up.