LLM

• AI Coding Agents Need a Source of Truth

  Bigger prompts do not fix bad agent plans. A good agent workflow starts with a small brief, human review, task sizing, and checks against concrete artifacts.

• How to Cook Claude Code

  Claude Opus 4.7 scores 87.6% on SWE-Bench Verified. Your daily experience probably doesn’t reflect that. The gap isn’t the model — it’s the kitchen. Three levers, a few hygiene habits, and the same model goes from autocomplete to autonomous shipping.

• It's Not the Model. It's the System Prompt You Never Wrote.

  Most complaints about coding agents are really complaints about empty context. CLAUDE.md, distilled project docs, and a few well-named slash commands turn the same model from ‘crappy autocomplete’ into something that runs unattended for hours and ships.

• I Asked ChatGPT If Building a Language Was Hard. Now I Have a Compiler.

  Four months ago I asked ChatGPT a dumb question while walking my dog. It told me building a programming language wasn’t that hard. So I built one — a compiled language with explicit error handling, generics, structured concurrency, and a garbage collector. Here’s what happened.

• When AI Agents Remember: Building Persistent Memory for Autonomous Coding

  What happens when you give an AI coding agent memory that persists across tasks? It stops repeating the same mistakes. Here’s how lgtm and snap use persistent findings, shared context, and self-recovery to close the loop from planning to merged PR.

• Scaling Up LLM Codegen Workflows for Bigger Projects

  A refined, step-by-step process for managing larger projects with LLM codegen workflows—response to Harper Reed’s blog post.