Ten durable guidelines for agentic coding with tools like Codex and Claude Code, covering specs, tests, taste, and maintenance costs.
Key Takeaways
Implement early and rebuild often; cheap code enables reconnoitering and thought experiments that were previously too costly.
Invest in end-to-end behavioral tests, not implementation tests, so you can freely rebuild without breaking contracts.
Keep specs (markdown goal files) updated continuously during implementation, not frozen pre-work artifacts.
Developer taste and domain expertise become the primary feedback loop when code ships faster than external feedback arrives.
Agentic code is “free as in puppies”: generation cost is near zero but maintenance, support, and security costs remain unchanged.
Hacker News Comment Review
Commenters broadly agree the real danger is executives misreading cheap code generation as cheap engineering overall, expecting headcount reductions that ignore maintenance debt.
A practical concern raised: once teams are fully dependent on AI-generated codebases they do not understand, reverting or auditing becomes prohibitively expensive.
One commenter flagged get-shit-done as a useful tool for enforcing upfront planning and keeping context small when using Claude, noting it trades speed for structure.
Notable Comments
@torben-friis: “Agentic code is free as in puppies” resonates as a sharp counter to LinkedIn posts celebrating millions of lines generated per month.
