Spec before code
Before writing code, produce an implementation plan for {feature}: files to change, data model, edge cases, failure modes, tests to add, and rollout risk. Ask me any blocking question first.💻 Profession hub
AI for Developers
Ship features with an agent pair, not just autocomplete.
Coding AI moved from completion to agents that plan, edit across files and run tests. The skill that matters now is specification and review, not typing speed.
Best AI tools for developers
What AI can actually take off your plate
Realistic automation levels for the tasks that eat the week — roughly 11 hours saved per week when the full stack is in place.
Boilerplate and scaffolding
90% automatable · 5 hrs/week
Test writing
80% automatable · 4 hrs/week
Code review first pass
65% automatable · 3 hrs/week
Documentation
85% automatable · 2 hrs/week
Prompts you can copy today
Refactor with rationale
Refactor this code for readability, performance and error handling. Output a diff plus a short rationale per change. Do not change behaviour; list any behaviour change you cannot avoid.
{code}Workflow: Agent-assisted feature delivery
Smaller diffs, better tests, fewer regressions.
- 1Write the spec prompt and review the agent's plan
- 2Let the agent implement in a branch
- 3Run tests and have the agent fix failures
- 4Automated PR review, then human review
- 5Merge with an AI-generated changelog entry
FAQs
Is AI-generated code safe to ship?
Only with tests and review. Treat agent output like a junior engineer's PR — useful, fast, and requiring verification.
Which model is best for coding?
Frontier models from OpenAI, Anthropic and Google trade places every few months. Pick the one wired into your workflow and switch when benchmarks move meaningfully.