Mapping How Developers Can Use AI

Work can be delegated. Responsibility cannot.

A navigation map, not a maturity model — use the level and context the work actually requires.

Understand System & Intent
Shape Change & Plan
Build & Integrate
Verify Quality & Security
Operate & Evolve
Explain a module, data flow, dependency, or design choice from the code and documentation I provide.
Suggest implementation options, edge cases, tradeoffs, and questions for a proposed change.
Complete, translate, or refactor a small code fragment while I remain in control of the implementation.
Suggest tests, review questions, failure modes, performance checks, and security concerns.
Explain logs, metrics, traces, or an error message and identify plausible causes to investigate.
Trace a request through selected services and repositories while I guide scope and correct its developing system model.
Turn a requirement into a design, work slices, migration steps, and rollback plan while I challenge each decision.
Edit files, run commands, resolve errors, and execute tests step by step under my active direction.
Build and run targeted tests, inspect failures, and revise the change with me until the evidence is sufficient.
Query logs, deployment history, infrastructure, and runtime state to investigate the incident path I specify.
Produce a system map for a feature, including components, data movement, dependencies, decision evidence, fragility, and unknowns.
Create a bounded change plan with architecture impact, alternatives, risks, migration, rollback, and unresolved decisions.
Implement a bounded change, update tests and documentation, and return the diff, assumptions, and verification evidence.
Independently review a change for correctness, security, performance, maintainability, and missing tests, with traceable findings.
Diagnose a production defect or performance issue and return evidence, likely cause, remediation options, and verification steps.
Maintain the system map as code, architecture, dependencies, decisions, and runtime evidence change, flagging assumptions that no longer hold.
When work becomes ready, create a change plan, identify risk, and request only the decisions or context that remain missing.
Implement approved, low-risk change classes from templates and policies, then open a pull request with evidence and rollback instructions.
Run review and verification gates on every change, blocking or quarantining results that exceed agreed uncertainty or risk.
Detect known anomaly classes, gather evidence, execute approved diagnostics, and open an incident or remediation proposal automatically.
Learn which relationships, decisions, and runtime signals best explain system behavior, improving the model while preserving versioned history.
Learn which planning gaps correlate with rework or failure and reprioritize analysis while preserving mandatory architecture and risk checks.
Improve playbooks, context selection, and tool choice from accepted, rejected, reverted, and defect-producing changes.
Tune test and review focus using escaped defects and false positives while retaining required controls and auditing suppressed findings.
Learn from incidents and remediation outcomes to improve detection, diagnosis, and response playbooks within explicit operational boundaries.

The Emerging Intent Layer

Codebase understanding and intent-driven development point toward the same missing capability: a persistent model of what the system is meant to do, how its parts relate, why decisions were made, and which constraints must remain true. Until implementation can be reliably derived from it and drift can be detected, it remains a governed system model rather than the sole source of truth.

Versioned intent & decisionsBehavior, constraints, rationale, and history evolve with the system.
Traceable to evidenceLinks requirements and architecture to code, tests, deployments, and runtime behavior.
Collaboratively evolvableTeams can inspect, challenge, edit, reconcile, and merge the shared model.
Drift-awareIdentifies where implementation or observed behavior no longer matches intent.

Guardrails Across Every Level

AI should improve understanding and delivery without becoming an unreviewed path for changing the software system.

Explicit authority & least privilegeAI may read, recommend, modify, merge, deploy, or remediate only within explicitly granted boundaries.
Isolated executionChanges run in reproducible, scoped environments before shared systems.
Evidence & traceabilityClaims, changes, tests, and decisions can be traced to their sources.
Independent verificationCompletion requires evidence beyond the implementing agent's own judgment.
Human override & auditA person can stop, reverse, inspect, or constrain any action or adaptation.
v1.0 · July 2026