Discover
FreeUnderstand the system, facts, limitations and first step.
STEP 1
Baseline, redesign, control, measure and prove a real task.
Identity, authority, permissions, approval, budget and shutdown.
Test claims, benchmarks, failure behavior, evidence and release status.
Measure spend and route work to the minimum sufficient method.
Data lineage, RAG, memory, privacy and freshness.
Incident replay, DR, offboarding, credential rotation and vendor exit.
STEP 2
What can go wrong? Is the action reversible? What is the financial, safety, privacy, legal or public-impact exposure?
What may AI recommend versus execute? Who approves? When does approval expire? Who can revoke it?
What must be sourced, retained, signed, auditable or explainable to a person later?
Budget per task, latency target, human-review budget, model/provider constraints and resource budget.
Human↔synthetic, synthetic↔synthetic, AI↔tool, or a combination? Syn-to-syn adds delegation, identity, TTL/depth/budget and loop-control requirements.
What must happen if the model, tool, agent, vendor, data source or human operator fails?
STEP 3
Understand the system, facts, limitations and first step.
Evidence, financial model, alternatives, failure modes and decision package.
Bounded configuration/implementation with tests, receipts, recovery and acceptance criteria.