Beyond 6 Months
A separate authority decision, not the reward for skills adoption.
The 6 Months Roadmap ends with a CTO investment decision, not automatic progression. The experiment may justify deeper improvement, selective replication, or stopping. The autonomous Target remains the long-term strategy vision; humans retain direction and review. The broad Roadmap connects skills and working practices to controlled autonomous capabilities. Six-month outcomes contribute toward that destination, but do not promise its timing or authorize its execution.
If one workflow works but transfer fails, invest in local context and coaching rather than mandate uniform use. If several teams request help and recurring value exceeds support burden, propose dedicated enablement capacity with the CTO and team leads. If deterministic tooling solves the problem more reliably, fund that. If outcomes remain inconclusive, narrow or stop the intervention. Recheck the next-quarter portfolio against business priorities, task supply, team willingness and ownership; do not extrapolate a pilot to all engineering.
Consider additional authority only when necessary
Carry forward proven skills where useful, measurements and working practices from real human + AI Agent work. The references provide governance, quality checks, adoption feedback, measurement and maintenance; none is permission to expand authority automatically. The diagram above applies only after a separate business case identifies an operation that supervised work or deterministic automation cannot adequately address.
Introduce authority deliberately
Establish reliable observation and evidence assembly, then proposing and routing work, before granting scoped execution authority. In the illustrative Delivery workflow, preparing a patch under human supervision does not authorize the central agent to merge or deploy it. A proposal must explain which extra operation is needed, why Stage 01 is insufficient, the exact resources/data it may touch and how humans can stop it.
The owning department's designated accountable approver decides the proposed workflow scope, with Architecture & Security accepting the authority/data boundary, Quality Engineering independently assessing behavior and Platform & IT verifying enforcement and recovery. These are recommended functions within existing teams—not appointed approvers or granted permissions. Unresolved release approval roles remain a hold condition; the agent cannot appoint its own authority.
Require exact-artifact approvals, separated credentials, independent checks, idempotency and duplicate/outage recovery, an auditable task ledger, and a tested stop/human-takeover path outside the model. Sensitive decisions remain human; missing evidence never means permission. A readiness failure returns to Stage 01 or redesign; it does not justify a larger experiment.
The output is either a recorded hold or a narrowly scoped authorization with conditions and invalidation rules. Monitor outcomes and authority violations separately. A material boundary change reopens the decision; harmful behavior revokes mutation authority and activates takeover, feeding evidence back into Quality and Maintenance.
This is future direction, not deployed autonomy, agreed delivery dates or approved KPI thresholds. The transition depends on reviewed evidence and explicit authority decisions, not adoption volume alone.
Return to the plan: 6 Months Roadmap.
Return to dependable work: Continue the Stage 01 improvement loop →.