Skip to content

Latest commit

 

History

History
27 lines (16 loc) · 2.32 KB

File metadata and controls

27 lines (16 loc) · 2.32 KB

AI Code Agents

  • Implementing a clear room Z80 / ZX Spectrum emulator with Claude Code

    • How to prompt a 'clean room' AI to follow instructions to make a complex project. A full spectrum 48k emulator!
  • agents.md

    • Think of AGENTS.md as a README for agents: a dedicated, predictable place to provide the context and instructions to help AI coding agents work on your project.
  • Why agents are bad pair programmers

    • give up on editor-based agentic pairing in favor of asynchronous workflows like GitHub's new Coding Agent, whose work you can also review via pull request

    • Continue to practice pair-programming with your editor, but throttle down from the semi-autonomous "Agent" mode to the turn-based "Edit" or "Ask" modes. You'll go slower, and that's the point. Also, just like pairing with humans, try to establish a rigorously consistent workflow as opposed to only reaching for AI to troubleshoot

    • My advice to the AI tool-makers would be to introduce features to make pairing with an AI agent more qualitatively similar to pairing with a human.

  • See [ai-local] for example of local code agent

  • Benchmarking GPT-5 on Real-World Code Reviews with the PR Benchmark

  • Vibe Coding as a Coding Veteran From 8-bit Assembly to English-as-Code

    • they are fully useful and safe only if you know what you are doing and are able to check and (re)direct what they might be doing — or have been doing unbeknownst to you. You can trust them if you can trust yourself.

    • And in this case, the gain in productivity is undeniable; at certain tasks it is easily 10X-100X in my opinion, like when you have to document the code, write unit tests, perform (not too complex) refactorings, write error messages, generate and capture exceptions, check the code for consistent and correct behaviour, code textbook or standard logic/algorithms/data structures, write a ton of boilerplate code, produce idiomatic source code… then rinse and repeat all of the above dozens of times…