n.V6-4.06 | WHERE THE AGENTS SIT IN THE WORK
01| Agents are not a department. They are threaded through the work, and the useful question is not "where is the AI" but "what verb does it own at each point".
02| Creating a project. The assistant helps a Maveriq draft the plan, surfaces comparable projects that already ran, and runs the compliance scan that gates progress. It does not generate the project. The Maveriq does. This is the single most important boundary in the book part, and it is easy to cross by accident: a system that produces a good first draft of everything eventually produces the ideas too, and then the barrio is executing the machine's imagination instead of its own.
03| Selling it. Agents run marketplace matching — relevance, not keywords: buyer thesis, geography, impact preference. They suggest a price range from comparable projects. Pricing is a hypothesis; the market corrects it. The buyer still chooses and the Maveriq still negotiates. AI shortens the search and surfaces signal.
04| Building it. Agents track progress, flag milestones at risk, escalate anomalies. The team executes. The agent is the early-warning system, not the project manager.
05| Measuring it. Agents process the long tail of data — adoption patterns, whether the impact held six months later, where things quietly evaporated. The community validates outcomes. AI measures; people interpret.
06| Learning from it. Agents aggregate ratings and find patterns across thousands of projects that no human reviewer would see. People still rate. AI extracts.
07| Training. This is the most fragile system we have, because our whole model depends on quality holding as Maveriqs teach the next generation of Maveriqs. Agents deliver curriculum in low-bandwidth and multiple languages, pace it to the individual, translate, run practice scenarios at unlimited volume, score the objectively scoreable, and match peers by complementary skill. They do not teach. The senior Maveriq stands in the room, because trust transfers through people, and because replication only works when the trainee can look at the trainer and see something they could become.
08| What agents add to training that humans cannot: drift detection. Each generation of peer-trained Maveriqs risks diluting the standard, invisibly, until the third generation no longer resembles the first. Monitoring that across a network is machine work. Correcting it is human work.
09| Read back the verbs. Prepare, surface, suggest, flag, track, aggregate, translate, monitor. Decide, teach, negotiate, resolve, validate — all human. That is not an accident of current capability. It is the design.
