“The code might have been messy, but the bigger issue was that the theory of the system, their shared understanding, had fragmented or disappeared entirely. They had accumulated cognitive debt faster than technical debt, and it paralyzed them.” margaretstorey.com/blog/2026/02...

It happened:

“What’s become clear: building software still demands discipline, but the discipline shows up more in the scaffolding rather than the code.” An example of the trend towards more architecture and feedback loops emerging from the trenches.

Read the (raw markdown) skill before you use it. If you use external skills this may be worth a watch. youtube.com/watch?v=Y2ot...

“We need to come together and think of Europe as a single market, with enterprises buying European technology, with states buying European technologies.” tech.eu/2026/02/11/m...

Those models I’m sure have their specific areas where they outperform the other ones but generally speaking Opus is not 3 times the value of Codex. I want to reduce wasted time fixing model errors as much as anyone else but Codex and Opus feel very close. (‘feel’: I didn’t do scientific study)

Be careful with environment variables in your project. Once an agent can run commands or code in your environment, “blocking” specific commands is meaningless. Here are some ways Reddit users reported how an agent circumvented its permission system that blocked read access to .env: 1/2

Every one of my agentic projects ends up with some custom setup of markdown docs for planning and task tracking. I’m trying hard not to call it too early, but this morning, I think I found the answer: Beads. It just feels like it's made for agentic work. github.com/steveyegge/b...

It's just a beautiful thing, having your tests ‘rolling green’ in `watch` mode when refactoring. It always has been, with LLMs I see them roll more often.

For those who dictate and sometimes have misses: I am using VoiceInk and adding Mistral's Voxtral (voxstral-small-24b-2507) as 'AI Enhancement' feels like a level up from built-in Parakeet V3. Parakeet was actually pretty good but decided to give Voxtral a go, I will not be disabling it soon.

Here is one of the dumbest things about vibe coding that I don't see a lot of people talking about Imagine you're an open source maintainer and someone opens up a pull request against your repository that they vibe coded. They may or may not tell you they vibe coded it; it doesn't really matter… 🧵

i made a version of wikipedia you can doomscroll xikipedia.org

Voxtral (from @MistralAI) transcription quality is quite incredible, the way it handles the punctation and all the rest, makes transcribed audio messages so much more understandable. I implemented a few fixes in the FFT and now there is no longer a skipped tokens issue in voxtral.c

Opus 4.6 and GPT Codex 5.3 in the same day. No coincidence but crazy times.

GPT-5.3-Codex is here. It advances both frontier coding performance and professional knowledge capabilities together in a single model. http://openai.com/index/introducing-gpt-5-3-codex/

LLM experience report: hanlho.com/p/experience... I built a personal AI assistant for my time tracking: I dictate in plain language, it edits my custom Markdown log and infers tags/projects. Used Agent Skills a building blocks. Coding agent this time: Codex. 1/2

Two new speech-to-text models (similar to Whisper) from Mistral today - one of them is API-only, the other is a 8.9GB Apache-2.0 licensed open weights model for "realtime" transcription. They're both very good! simonwillison.net/2026/Feb/4/v...

The CIA just stopped publishing their World Factbook and took every page, including the archived copies of previous versions! This sucks. It was public domain, so I recovered the 2020 edition (the last one published as a zip file) and shared it to GitHub simonwillison.net/2026/Feb/5/t...

New blog post: AI’s Opportunity: Pacing Control Loops with Development hanlho.com/p/ais-opport...

AI doesn’t remove quality practices; it raises the stakes if you don’t have them

LLMs and coding agents are a major opportunity to adopt best development practices and integrate them directly into workflows. Code reviews seem the easiest to start with: low barrier of entry, often valuable insights.

Lucky, just switched from Claude Code to Codex Feb 1st

With every skill becoming also command across coding agents quickly, typing “/” gets unwieldy fast. I like Codex’s split: use “$” for skills/custom commands only; “/” should only show its built-in commands.

New post: Code quality feedback loops in AI dev workflows Nothing exciting, just some basic ideas I've been applying, but it helps me to understand where AI-assisted development is likely heading. nick-tune.me/blog/2026-02...

“45 tips for getting the most out of Claude Code, from basics to advanced” github.com/ykdojo/claud...

I wrote about Clawdbot/Moltbot/OpenClaw and Moltbook, the fascinating, weird and sometimes even useful social network for digital assistants to swap tips and gossip with each other simonwillison.net/2026/Jan/30/...

Wrote a short post on using LLMs in relation to general software engineering best practices. Working with LLMs requires more discipline and attention to fundamentals, not less. hanlho.com/p/on-buildin...

My ask: ✅ Download the doc ✅ Start conversations with your dev & security teams about what to do, not just what to avoid. 🙏 https://twp.ai/Imtgl2 5/5

Mistral Vibe 2.0 got released. Catching up quickly. Included in Le Chat Pro subscription. ‘Generous usage limits.’ mistral.ai/news/mistral...

"Subscribe with Bluesky" (to a blog) A bit much, not?

My Claude Code subscription runs out in a few days, so I’ll be relying even more on Opencode. This interview adds useful background about Opencode and its team. Found it reassuring, especially around how it differs from Claude Code and their planned business model: youtu.be/kKGeOE641b0?...

By lack of a better word but it's the first word that comes to mind: brilliant article by Chad Fowler on relocating rigor when using LLM agents. aicoding.leaflet.pub/3mbrvhyye4k2...

Qwen released a new Apache 2.0 licensed text-to-speech model, with full voice cloning abilities, and it's really effective - my notes from trying it out here: simonwillison.net/2026/Jan/22/...

Today’s goal: write a blog post about the vibe-coding experience writing spelling games for my daughter. Today’s outcome so far: three Opencode sessions, zero paragraphs.

i didn't think this would be a big deal or happen so fast but i'm seeing teams nerf their own ability to use their brains because of llm dependence and when they run into a problem the llm can't fix they start doing really weird stuff

Too many models on Opencode cleaned up: (Snapshot, will probably be different in a month.)

Once again, the buy #European is trending. What are your favorite European brands? 🇪🇺 In our latest guide, we take a look at the best European products you can use 👉️ tuta.com/blog/boycott... #BuyEuropean #PrivacyFirst #GDPR #DigitalSovereignty #PrivacyTech

Last week Claude Code refused to create a simple custom command because: 'There's no separate "command" concept - skills are what provide the /command-name interface.' So I wrote a little piece on it: hanlho.com/p/the-unific...

‘Dear America, We'd Like to Speak to the Manager’ youtube.com/watch?v=dBoL...

Weekend thoughts on Gas Town, Beads, slop AI browsers, and AI-generated PRs flooding overwhelmed maintainers. I don't think we're ready for our new powers we're wielding. lucumr.pocoo.org/2026/1/18/ag...

I like watching Nick’s agentic coding evolution ‘live’ sort to speak, interesting stuff. It seems like this is moving to a setup where we explicitly have to decide where we think an LLM is supporting us reliably enough and where we want keep control. And put the latter in code and/or scripts.

“Don't be evil”

So this morning the Claude Code plan is again blocked from Opencode. Anthropic is doubling down on this one. What is interesting is that they specifically seem to be targeting Opencode. For example Droid and Cursor do still work. xcancel.com/ryanvogel/st...

Wow. +10K in 6 days. #opencode

Signal creator Moxie Marlinspike wants to do for AI what he did for messaging arstechnica.com/security/202...

Confused about all the names test objects can have? Here is a clear explanation of existing terms and some history. With a useful recommendation: "start talking about what these [test] objects do for you instead of what they are called. Use verbs."

Published: What is Continuous Delivery? It takes nine principles, one heuristic, three foundations, one pattern and three practices to practice truly Continuous Delivery. thinkinglabs.io/articles/202...

Sharing my LLM agent skills and workflows in a new repo: Agent Chisels. Includes skills for ADR documentation, Jujutsu (jj) version control, and a meta-skill for evaluating skills. Blog post: hanlho.com/p/agent-chisels-my-llm-agent-skills-and-workflows/ Repo: github.com/lhohan/agent-chisels

New blog post: Don't fall into the anti-AI hype. antirez.com/news/158