Methodology

Goal → evidence → gap → project → tool

The planner uses transparent rules. It does not call an AI model or upload responses.

Goal first

The 90-day result determines the context.

Evidence over confidence

Observable work determines the stage.

Reliability before novelty

Testing and governance outrank fashionable tools.

Smallest credible tool

Use the least complex route that can deliver the result.

Risk-sensitive guidance

Recommendations change for sensitive data.

Bounded agent readiness

Persistent agents require operational prerequisites.

Capability stages

  1. Explore — Use AI for individual tasks.
    Examples: Research briefs, drafting, ad-hoc queries, code snippet explanations.
  2. Apply — Create reusable instructions and resources.
    Examples: Custom prompt templates, system instructions, team-wide style guides.
  3. Build — Produce a working assistant, workflow or application.
    Examples: n8n/Make automation flows, Claude Code scripts, custom web app prototypes.
  4. Validate — Test expected and actual behaviour.
    Examples: Evaluation test suites, LLM-as-judge benchmarks, edge-case regressions.
  5. Integrate — Connect real inputs and systems.
    Examples: API integrations, database RAG pipelines, CRM webhooks, multi-channel gateways.
  6. Deploy — Add users, controls, monitoring and recovery.
    Examples: RBAC permissions, audit logging, fallbacks, cost/token monitoring dashboards.
  7. Lead — Prioritise and measure AI across an organisation.
    Examples: Resource allocation, AI governance policy, team capability roadmaps, ROI metrics.