One person, both halves.
The person who diagnoses the workflow is the person who builds the system. Nothing is lost in a strategy-to-developer handoff.
The work moves from diagnosis to evidence, then through explicit knowledge, data, permission, and release gates before a system reaches operation.
Strategy and implementation stay with the same accountable person throughout.
Nine stages in three phases. Each phase closes before the next one is allowed to start.
Diagnosis, mapping, a value hypothesis, and a tested prototype come before any production decision.
Start with the costly, inefficient, or strategically important workflow. The right recommendation may be a system, an integration, a specialized tool, or not to build when the value is not there.
Trace the workflow, recurring decisions, handoffs, knowledge sources, and available baseline before choosing technology.
Identify where the workflow may create commercially meaningful leverage, then measure the available baseline or establish a measurement plan.
Build a working prototype, test it with representative users where access permits, and use the evidence to define the production recommendation.
Knowledge, data, and permission boundaries are written down before the system is allowed to act.
Define what the AI may read, what stays inside client-controlled systems, what it may draft or propose, and what remains prohibited.
Use explicit permissions, human approval, provenance, evaluations, and release gates. The model is never the permission system.
The system ships as an owned, documented, model-portable asset and is rolled out to the team.
Deliver the logic, rules, workflows, evaluations, integration code, configuration, documentation, and repository to the client.
Design the business logic so a provider change does not force a rebuild from zero. Provider changes may still require adaptation and retesting.
Deploy the validated system, document it, support basic team rollout, and expand into more roles or workflows when justified.
The gate is the center of the system: the model can reason and propose, but authority remains explicit, human, and verifiable.
The model is never the permission system.
A visible gate separates analysis from action. A proposal waits at the boundary until a person approves it; the result is then checked and returned to the record.
They hold for every engagement, whatever the workflow and whichever model provider the system uses.
The person who diagnoses the workflow is the person who builds the system. Nothing is lost in a strategy-to-developer handoff.
You receive a version-controlled repository containing the system logic, rules, workflows, evaluations, integration code, configuration, and documentation. Sensitive data, private records, credentials, and operational knowledge sources remain inside systems you control.
Every system defines what AI may read, draft, or propose; what a human must approve; what remains prohibited; where important answers came from; and how actions are verified.