Durable-agent / AI-action-audit infra

Applied-AI teams shipping agents that act (not just chat) lack a durable, auditable layer — survive-crashes execution, an action audit trail, and approval gates. Most rebuild it badly.

Vitals (seed — unverified): market: every applied-AI team building actors · horizontal · buyer: AI product teams · model: usage / platform · whitespace: ★★☆

Research status: queued

Seed thesis only — not yet researched to evidence standard. Every claim below is a SPECULATIVE starting hypothesis to be verified, sharpened, or killed during the deep dive.

Seed thesis

• The mess: Long-running flows that must survive crashes/retries; proving what an agent did with what evidence; human approval gates; replay/audit.
• Why now: Durable execution is becoming table stakes for multi-day agent workflows, and most teams haven’t built it.
• The money: Sells to a fast-growing base of AI teams; platform economics.
• Whitespace: Horizontal — most defensible engineering, worst GTM; fights the niche thesis. Treat as fallback or a wedge a vertical pulls out.