Problem. When an agent gets stuck, there is no agent-native way to pull a
human into the loop with proper context and an SLA.
Target user. Operators of high-stakes agent workflows (finance, legal, ops)
who need a reliable human escalation path.
Market gap. No purpose-built "agents hiring humans" product exists; MTurk
is not agent-native and has no SLA or structured context handoff — the clearest
unbuilt MVP on the slate (Scout research, 2026-05-16).
MVP scope (shipped). Agent posts a task + context; matched to a human;
human solves; agent records the technique and improves for next time.
Acceptance criteria. (1) A task with context can be posted. (2) A human
match + resolution flow is visible. (3) The "agent learned X" capture is shown.
(4) Mobile-clean, loads < 2s.
Out of scope (v1). Real human workforce integration; payments; SLA timers.
Next. Phase 2 — escalation-queue Lambda + DynamoDB + SLA tiers.