The third rung of the Boris Cherny's 4-Step AI Maturity Framework, following Step 1 – AI Assisted (Pair Programmer Model) and Step 2 – AI Builds in Parallel (Orchestrator Model). At this stage Claude stops waiting to be told what to do next: it proactively starts work and runs maintenance/cleanup continuously in the background, without being manually kicked off.
"At this point, Claude proactively does work that you would have had to kick off manually before. Maintenance and cleanup that used to wait for someone to find the time now runs continuously in the background."
The corresponding shift on the human side is a mindset change: from doing tasks, or even orchestrating tasks, to managing an org chart of AI "employees" — an "AI agency CEO" posture. Reaching this step depends on having already solved verification (Rule-Based (Objective) Agent Output Verification, Taste-Based (Subjective) Agent Output Verification) and permissioning ('Drunk Intern' Agent Permissioning Model), since autonomy without either is what makes unsupervised agents dangerous rather than useful.
Practical mechanisms the presenter uses to run this "AI org" are claude-skills-as-sops (SOPs), claude-routines-task-scheduling (recurring cadence), ai-agent-reporting-loop (status updates), and sub-agents-parallel-fanout (fanning one task across multiple agents).
Apply: Stop thinking task-by-task and start managing AI the way you'd manage a growing team — accept the short-term setup cost (writing SOPs, wiring up reporting) for long-term leverage, per eighty-percent-good-enough-delegation.