QConsul LLC — a certified Oregon Benefit Company. Portland, Oregon, USA.
Agent Lifecycle Pipeline — from specification to live operation and ongoing refinement
A case study of QConsul's 18-step, 4-lane pipeline (Operator, Agent / AI, Governance, Platform) taking an agent from spec.md to live operation to retirement, with an explicit HITL / HOTL / HOOTL oversight-mode lens on Steps 12 / 13 / 15 / 16 / 18, and a real 2026 incident where the HOTL control caught what monitoring would not have.
Published July 16, 2026 · by Karen Michael, Founder, QConsul LLC.
Worked example: QubieClaw / @LambdaSays, June–July 2026
QubieClaw runs its Moltbook engagement cycle explicitly without HITL — no pre-publish human review. The compensating control is HOTL: periodic human review with a 30–60 minute intervention SLA. A peer agent's adversarial critique flagged 'conviction hallucination' risk on a live post; the critique was caught on periodic human review (not automated telemetry), fed into the 2026-07-07 self-assessment, and produced a citation-grounded follow-up post directly addressing the risk. Live demonstration that 'no HITL' did not mean 'no one could catch it.'
Frequently asked questions
What is HITL, HOTL, and HOOTL?
HITL (human-in-the-loop) requires human review or approval before each action. HOTL (human-on-the-loop) lets the agent act autonomously with periodic human review and a defined intervention SLA. HOOTL (human-over-the-loop) lets the agent act fully autonomously within pre-approved boundaries, with humans reviewing aggregate outcomes and holding kill-switch authority. Mode is assigned per risk tier at Step 12 of the pipeline.
How can a no-HITL agent still be governed?
By running HOTL as the compensating control: periodic human review of outputs and external signals, with a defined intervention SLA and a kill switch at Step 18. The QubieClaw worked example shows a live no-HITL agent whose 'conviction hallucination' risk was caught on periodic human review of an adversarial peer critique and corrected through a citation-grounded follow-up post.