Deleting CLAUDE.md, Re-Tested 6 LLMs, and Gemini 3.1 Pro
And more AI Coding news: February 25, 2026.
Hey hey,
Another week full of AI coding experiments. It’s never really simple, with all those models, IDEs, tools, workflows and recommendations keep changing. So I’ll keep experimenting, join me by watching the videos below!
For Paid Substack Subscribers
I Re-Tested 6 LLMs on 6 Laravel Projects: 3 Takeaways
13-minute video: version 2 of my Laravel Evals for AI coding models, with 6 new projects and challenges. One clear winner, and, surprisingly, it’s not Opus model.
My YouTube videos
I Tried Deleting CLAUDE.md. Here’s What Happened.
I’ve seen some people recommending to shorten or even delete CLAUDE.md or AGENTS.md files. I’ve tried it myself, on 5 different Laravel projects.
I Tried Gemini 3.1 Pro in Google Antigravity
I tested the new model with a few Laravel projects, let me show what happened.
[Short] Claude Code: Change Model While Writing Prompt?
Writing a long prompt and forgot to switch the LLM? This tip may be very helpful.
[Short] Prompt to Automate Some CLAUDE.md Rules
Not everything needs to be in CLAUDE.md.
Website Owners Will/Should Serve AI Agents Differently?
A quick video about emerging new trend, to work with AI agents as website clients more efficiently.
From AI Coding Community
Lessons from Building Claude Code: Prompt Caching Is Everything.
x.com
Long article on X by Thariq from Claude Code.
Start local sessions from the terminal, then continue them from your phone. Take a walk, see the sun, walk your dog without losing your flow.
johannesjo/parallel-code
github.com
Run Claude Code, Codex, and Gemini side by side — each in its own git worktree
I’ve published the first two chapters of a new guide to Agentic Engineering Patterns
x.com
By Simon Willison - coding practices and patterns to help get the best results out of coding agents like Claude Code and OpenAI Codex.
VS Code extension that gives each Claude Code agent its own animated pixel art character in a virtual office. Free, open source.
Codex 0.102 is out with experimental multi-agent support.
x.com
TUI: enable it under /experimental - multi agents
Claude Code hackathon results
x.com
500 builders spent a week exploring what they could do with Opus 4.6 and Claude Code. Meet the winner projects.
Introducing Claude Code Security, now in limited research preview.
x.com
It scans codebases for vulnerabilities and suggests targeted software patches for human review, allowing teams to find and fix issues that traditional tools often miss.
Anthropic sent a legal notice to OpenCode
x.com
They sent a legal notice to OpenCode, making official what many have feared (while not making an official statement). External tools are now explicitly prohibited from using or even orchestrating claude code.
No more re-explaining what you were doing, or painful copy/paste from terminal that lose tool-call context.
Jarred Sumner on X: “Claude Code’s p99 memory usage dropped by 40x in the last two weeks”
x.com
And by 6x since January - while growing rapidly and continuing to ship tons of new features. So, if your Claude Code feels slow, look at this chart and at the version you’re using.
That’s a wrap for this week. Keep building with AI!
Povilas Korop
AICodingDaily.com
The premise misses the point. The question isn't "should you have a CLAUDE.md?" but "what belongs in it versus elsewhere?"
I run 200+ condensed lessons across 10 topic files with auto-rotation when the main file exceeds 60 lines. Not because bigger is better, but because those 200 lines represent 400+ hours of calibration that compounds. Every lesson prevents an entire class of errors from recurring.