1
Assess
Map the current AI-native stack: data readiness, context access, existing agents, interfaces, and governance gaps.
AI development loop
Build the right things, then make them work.
A macro loop for direction.
A micro loop for quality.
A micro loop for quality.
Operating cadence
↓
2
Discover Use Cases
Combine technical feasibility with organizational reality. Pick a bounded set of high-value, low-friction candidates.
↓
Micro loop: locked scope, fast iteration
3. Deploy
Build the selected agents, wire the tools, and train the humans who will manage the workflow.
4. Evaluate
Measure performance, gather feedback, inspect failures, and refine the same scoped deployment.
↓
5
Retrospective
Step back from delivery. Decide what the organization learned, what should change, and where the next cycle starts.
Separate Direction From Execution
Assessment and discovery answer whether the organization is building the right things. Deploy and evaluate answer whether those things are being built well.
Discovery Needs Two Kinds Of Knowledge
AI engineers understand what agents can do. Domain experts understand where value, friction, and cultural constraints actually live. The process creates overlap.
Use Two Tracks
Deploy foundational use cases quickly while interviews and workshops surface deeper opportunities for later cycles.
Foundation
Start from reusable use cases that build literacy and create immediate working assets.
Specificity
Mine interviews, workshops, and process data for opportunities unique to the organization.
Optionality
Each cycle leaves the organization with clearer capabilities, better context, and a ranked next set of bets.