Use cases

AI does most of the work. Humans perform the critical part AI should not do alone.

Every use case below follows the same division of labor: the agent does the bulk of the task, identifies its own uncertainty or risk, and requests a small, well-scoped human intervention. The human result returns to the agent, and the agent continues.

AI coding agent requests a senior architecture review

An AI coding agent designs a service and generates most of the implementation. Before committing, it detects that the architecture decisions are high-consequence: the wrong choice costs weeks.

• AI does: draft architecture, generate code, enumerate trade-offs.
• Human does (ARCHITECTURE_SANITY_CHECK): review the architecture for critical issues, scaling risks and recommended changes.
• Result: structured verdict and findings flow back to the agent, which adjusts the design and continues.

Founder verifies only the uncertain clauses of a contract

A founder uses AI to analyze a contract end to end. AI summarizes everything, but flags a few clauses where legal wording is genuinely uncertain.

• AI does: full contract analysis, clause extraction, identification of uncertain clauses.
• Human does (future LEGAL_EXPERT_CHECK): review only the flagged clauses, not the whole document.
• Result: verdict and explanation for each flagged clause; the founder proceeds with high confidence.

Research agent verifies one critical assumption

A research agent builds an analysis on an assumption it cannot verify with high confidence from sources alone. The assumption is load-bearing: the whole conclusion depends on it.

• AI does: research, synthesis, identification of the critical unmet assumption.
• Human does (EXPERT_FACT_VERIFICATION): verify the claim, with verdict, explanation and source notes.
• Result: the agent either proceeds or revises its conclusion based on the verdict and confidence.

Physical-world verification that cannot be performed digitally

An agent needs to confirm something that exists only in the physical world: an asset's condition, a site state, a material fact.

• AI does: plan the check, specify what must be verified and what proof is required.
• Human does (future PHYSICAL_VERIFICATION): go to the location and gather proof (photos, GPS, timestamps).
• Result: verifiable, structured evidence returns to the agent.

The principle

In every case the human's task is small, well-scoped and critical. This keeps expensive expertise affordable in small units and keeps the agent in control of the workflow. The agent requests an outcome; the platform arranges the human.