Back to roles

AI playbook for product engineering

Developers use AI to move from requirement to reviewed code faster

Developer teams pair human design judgment with AI assistants for discovery, implementation, refactoring, tests, and pull request preparation while keeping architecture, security, and code ownership human-led.

92% of product squads use AI weeklyAbout 210 hours saved per sprint100% of pull requests can benefit from AI assistance

Developer tool stack

AI assistants embedded in the engineering loop

Developers keep AI close to the IDE, repository, and review flow so repetitive implementation work gets faster without weakening engineering judgment.

Claude Code

Used daily

Default coding agent for repository-aware exploration, multi-file changes, refactoring plans, and implementation support with reviewable diffs.

GitHub Copilot

Used daily

IDE-native assistant for code completion, boilerplate generation, test scaffolding, and quick explanations while working inside the editor.

ChatGPT

Used daily

Reasoning workspace for clarifying requirements, comparing solution approaches, drafting prompts, and explaining unfamiliar APIs or patterns.

Cursor

Used weekly

AI-native editor used for context-aware refactors, codebase navigation, and fast iteration on feature branches.

Spec-Driven Development

Used weekly

Structured specs that align product intent, implementation tasks, tests, and review criteria before AI starts changing code.

Ticket to pull request

AI-enhanced developer workflow

A developer playbook that keeps humans responsible for intent, architecture, validation, and merge decisions.

  1. 1

    Understand the ticket

    What we do

    Read the requirement, inspect related code, identify constraints, and confirm the expected behavior with product or QA.

    How AI helps

    Summarizes the ticket, turns acceptance criteria into an implementation checklist, and highlights unclear edge cases.

  2. 2

    Plan the change

    What we do

    Choose the technical approach, define touched modules, evaluate risks, and decide what tests are needed.

    How AI helps

    Suggests implementation options, maps likely files, compares trade-offs, and drafts a step-by-step plan for review.

  3. 3

    Implement incrementally

    What we do

    Write or review code in small slices, keep conventions consistent, and protect existing behavior.

    How AI helps

    Generates boilerplate, proposes focused patches, explains unfamiliar code, and helps refactor without broad churn.

  4. 4

    Test and debug

    What we do

    Run unit, integration, and manual checks; inspect failures; and verify the change against acceptance criteria.

    How AI helps

    Drafts unit tests, generates test data, explains errors, and proposes debugging paths based on logs or stack traces.

  5. 5

    Prepare the pull request

    What we do

    Review the diff, remove noise, document decisions, and make sure the PR is understandable for reviewers.

    How AI helps

    Creates PR summaries, identifies risky files, suggests review notes, and checks whether tests cover the intended behavior.

  6. 6

    Learn and reuse

    What we do

    Capture useful prompts, patterns, and lessons so the next sprint starts with better examples.

    How AI helps

    Turns completed work into reusable snippets, coding guidelines, and onboarding notes for the team.

Real development moments

Where AI helps developers every sprint

The highest-value use cases cluster around clarity, implementation, testing, and maintainable handoff.

Requirement and design clarity

  • Convert vague tickets into acceptance criteria, edge cases, and questions for product.
  • Compare implementation paths before committing to a design.
  • Draft spec sections that define behavior, constraints, and test expectations.

Implementation acceleration

  • Generate CRUD services, DTOs, GraphQL schemas, API handlers, and validation code.
  • Refactor legacy modules with explicit constraints and existing conventions.
  • Translate patterns from one module into another without copy-paste drift.

Testing and debugging support

  • Create unit test and integration test scaffolds from acceptance criteria.
  • Generate edge-case datasets for validation, permissions, and failure paths.
  • Explain stack traces and narrow the likely source of regressions.

Review and documentation

  • Summarize pull requests with user impact, technical decisions, and test evidence.
  • Rewrite comments and docs so future maintainers understand why code exists.
  • Create release notes or handoff notes from implementation context.

Field-tested prompts

Prompts developers can reuse

These prompts keep context explicit and make AI output easier to review.

Scenario

Clarify a ticket

Turn this ticket into acceptance criteria, edge cases, and clarification questions: ```[ticket]```

Scenario

Plan a focused change

Given this requirement and repository context, propose a minimal implementation plan with files to inspect first: ```[context]```

Scenario

Generate tests

Create unit tests for this function using the existing project style. Include edge cases: ```[code]```

Scenario

Explain a failure

Explain this error, list likely causes, and suggest the next debugging commands: ```[stack trace or log]```

Scenario

Refactor safely

Refactor this code for readability without changing behavior. Explain each change briefly: ```[code]```

Scenario

Review a diff

Review this diff for bugs, missing tests, security risks, and unclear naming: ```[diff]```

Scenario

Write a PR summary

Write a concise PR summary with behavior changes, test coverage, and reviewer notes: ```[diff or notes]```

Scenario

Document a pattern

Turn this implementation into a reusable team guideline with examples and anti-patterns: ```[code or notes]```

Best practices

Do and don't guidelines for AI-powered development

Do

  • Start from the requirement and repository context before asking AI to write code.
  • Ask for small, reviewable changes instead of broad rewrites.
  • Run tests and inspect diffs manually before merging.
  • Keep architecture, security, and product decisions owned by humans.

Avoid

  • Do not paste secrets, customer data, or private credentials into prompts.
  • Do not accept generated code that you cannot explain.
  • Do not use AI output as a substitute for code review.
  • Do not let AI introduce new dependencies without a clear reason.

Risks and mitigation

Keeping AI-assisted engineering safe

Confident but wrong implementation

Risk #1

What happens: AI can produce plausible code that misses domain rules or breaks hidden assumptions.

Mitigation: Validate against acceptance criteria, run focused tests, and ask reviewers to inspect behavior-critical paths.

Security and data leakage

Risk #2

What happens: Prompts can accidentally expose secrets, customer data, or internal implementation details.

Mitigation: Sanitize inputs, use approved tools, and keep sensitive data out of external prompts.

Over-refactoring

Risk #3

What happens: AI may broaden a small task into unrelated cleanup that increases review cost and regression risk.

Mitigation: Constrain scope, request minimal patches, and reject unrelated churn before opening the PR.

Skill erosion

Risk #4

What happens: Developers can lose debugging and design intuition if AI answers are accepted too quickly.

Mitigation: Require engineers to state hypotheses, explain generated code, and compare alternatives before merging.

What comes next

How developer AI maturity evolves

Immediate enablement priorities

Junior developers need AI to accelerate learning without hiding fundamentals.

  • Prompt templates for reading code, writing tests, and explaining errors.
  • Mentored review rituals where engineers explain AI-generated code before merge.

Future-state engineering leverage

Senior developers use AI to scale architecture thinking, quality review, and reusable patterns.

  • Spec-driven workflows that connect requirements, code changes, tests, and PR review.
  • Repository-aware agents with strong guardrails, audit trails, and team-approved standards.

Ready to bring this developer playbook into your organization?