Takes a task from description through implementation, testing and pull request with limited supervision.
Works a task end to end: understanding the requirement, exploring the codebase, implementing, testing, iterating on failures, and opening a pull request. The most capable and least predictable solution in this group.
Well-specified bounded tasks in codebases with good test coverage — the coverage is what makes autonomy tolerable.
Ambiguous requirements, poorly tested codebases, and anything where the agent's own tests are the only verification.
The agent's changes go through the same review and approval as any other. Its commits must be attributable to it, not to the human who invoked it, or the change record becomes false.
Working around a failing test rather than fixing the defect it detected — the agent optimises for the tests passing, which is not the same objective.
Implementing a new file-format parser against an existing interface and test harness, where the specification is precise and the tests are not the agent's own.
Two different absences share this shape. Foundational solutions get built whatever the domain, so no domain links them; the rest are solutions this domain genuinely does not reach for. v_ai_solutions_unlinked separates the two.