Agent Action Reversibility-Window Council for In-House Commerce Agents

An async council API for commerce operators running in-house LLM agents (custom GPT/Claude integrations against Stripe, Klaviyo, Postmark, Gorgias API) that POST proposed actions to AE before commit, OR forward post-action webhooks from semi-open platforms inside the reversibility window. AE runs the multi-model adversarial council (30-60 seconds, not sub-second) against the action JSON, returns approve / require-human / reverse with named failure-pattern reasoning, and for post-hoc cases issues the reversal API call (Stripe refund reversal, Klaviyo campaign pause, follow-up correction message) before damage settles. Sold to 10-50 person DTC ops leads who already have one engineer maintaining the in-house agent loop — not to brands using only closed Decagon/Sierra. Stripe chargebacks and Gorgias escalation tickets close the 14-30 day reality-graded loop. The 6-pattern autopsy (Cosmetic Confidence, Fatal Grounding Immunity) names exactly what slipped past the agent's self-check, feeding pattern-strength updates back into the council weights.

AE's adversarial council, autopsy taxonomy, and reality-graded loop fit the post-action reversibility window of in-house DTC commerce agents better than any sub-second chatbot platform, and the ICP is small enough for a solo UK operator to reach £100–300K ARR.

Still in evaluation (phase: ranked). No verdict yet.