5 Stage Gates

The first gate confirms that the problem definition is complete and shared before any specification work begins. A vague or contested intent definition makes every downstream artefact unreliable specification, planning, execution all become noise built on a shaky foundation. The Product Owner holds this gate because they are accountable for whether the right problem is being solved.

Specification quality is the single greatest determinant of agent execution quality. An agent given vague requirements will produce plausible but incorrect output. The Tech Lead holds this gate because they are best positioned to assess whether the acceptance criteria are precise enough to be machine-checkable, and whether constraints and edge cases have been fully surfaced before planning begins.

Before the agent enters execution, the plan must be reviewed and the operational guardrails explicitly set. The Sponsor holds this gate because authorizing a plan commits resources, sets the risk envelope, and locks in the guardrails that bound what the agent may do autonomously. This is the last opportunity to adjust scope and constraints before execution begins at scale.

The verification gate is not a rubber stamp on test results it is a human review of the agent's decisions, the evidence package, and the risk posture before anything ships. QA holds this gate because they are uniquely positioned to assess whether the evidence of correctness is genuinely sufficient or merely superficially complete. Agent self-assessment is necessary but not sufficient for release.

The final gate authorizes production deployment. This is the accountability moment where a named human the Engineering Manager accepts organizational responsibility for shipping. Monitoring must be configured, rollback must be rehearsed, and deployment sequencing must be confirmed before authorization is given. The agent suggests the release plan; the human authorizes it.