📝 My contribution to the discourse on the software industry's favorite topic du jour. seanvoisen.com/writing/prog...

Recently, I put all of my team's first-principles thinking about the future of computing into one demo. It's our vision of the capabilities a truly AI-native personal computer will have. ruperts.world/blog/ai-comp...

I find it funny when Claude tries to talk me out of building something like "you _could_ build it that way but it would probably take a month" and then builds it in a hour

Atelier started as a set of custom agent skills that I've been iterating on as I moved from chat-based development to spec-driven development. The kanban board is how I organize it all, but it's just a window onto the workflow.

No tool is magically going to save you from the technical debt of vibe coding. Verifiability has to be baked into every step. Atelier attaches specific acceptance criteria to every task and automatically verifies them.

One of my favorite little details is that you can hover over blocked tasks to quickly see the blocking tasks or hover over a plan to see which subtasks are completed

The new Agent Teams feature in Claude Code lets multiple agents work simultaneously. Fun watching the atelier.dev cards update automatically as each finishes: code.claude.com/docs/en/agen...

I added a direct link to the .vsix file for Cursor users until atelier.dev is available in the Cursor extension registry

Going fast with AI is less about generating more code. It’s more about gardening a working system whose complexity scales linearly as it grows.

I built Atelier because I needed a better way to organize specs for my agent workflows. When I click Implement on a task, it runs Claude Code with my custom prompt template. When Claude finishes, the card automatically moves to review. youtu.be/eVZJc7Rz8OI

I put a lot of work into this release 😊 excited to share the new features! www.spacecake.ai #ClaudeCode #opensource

Would love to see this get widely adopted. Having deep links between code and conversation history would be so valuable for both developers and agents agent-trace.dev

Under the hood, Atelier uses Claude Code's new task tools and reads the Markdown plans from planning mode. The spec is just a file. The tasks are just files. Everything stays editable and actionable.

Here's my new workflow: I write a short todo. Claude helps me turn it into a detailed spec, then into a task list, then works through the list. I check the kanban board, verify what's done, course-correct if needed.

I wanted to explore if we can build something like Claude Cowork in the browser that can work with the user's file system safely running programs against them. We can get a very long way: aifoc.us/the-browser-... Feedback and critique appreciated.

Looking forward to reading Maintenance: Of Everything by @stewartbrand.bsky.social How cool is this kintsugi cover

New flavor of imposter syndrome: I haven’t even used all my Claude Code quota today

5 years ago I switched to an ultrawide computer monitor and it’s been one of the best decisions I made. So much of my workflow these days starts as a single thread and then expands into 4 or 5 columns of side-by-side documents, AI agents and custom tools…

prototyping co-drawing with Gemini Flash 3 at Google in these demos "thinking" is disabled, which makes the model return tokens very quickly (all videos are realtime), and I find these rapid responses pretty good for the use-cases I'm experimenting with, like: executing simple diagrams ...

Wrote a blog post about the token encoding format in GPT tokenizers. Have you ever inspected a tokenizer and seen tokens like оже or ĠнÑĥжно? These are tokens in a custom encoding format used for serialization of byte-level tokenizers. This post shows how to recover the original token.

Introducing Splat 🟨📌: Affinity diagramming in a single HTML file. Drag notes and organize ideas on a visual board. With semantic search and an agentic AI assistant. Open source and fully local. Check it out here: github.com/ianarawjo/sp...

“If understanding lives in prompts and specifications rather than implementations, what happens to the craft of programming? … The definition of quality may be moving from "durability" to "regenerability"—from code that lasts to code that can be reliably reproduced.”

Love this vision for sensemaking computers

Claude Code is starting to write out functional requirements as part of the planning process www.anthropic.com/engineering/...

FIles are the new database -- most codebases are already well structured, why not aid the coding agent with a file based index at the root of each folder, generated in seconds using TreeSitter? Runs as a skill, invoked on demand. Supports JS, TS, Ruby, Go, Rust, Java, and of course Python.

Putting on my noise cancelling headphones so I can listen to some high quality noise

I’m seeing confusion about Claude Skills and why you need them. It sort of feels like Anthropic is building tools that fix problems many people won’t know that they have until next year www.anthropic.com/engineering/...

I’m seeing a lot of AI coding tools trying to figure out multi-agent UX. To me, delegating feels like the least interesting part of the problem…

Do AI agents ask good questions? We built “Collaborative Battleship” to find out—and discovered that weaker LMs + Bayesian inference can beat GPT-5 at 1% of the cost. Paper, code & demos: gabegrand.github.io/battleship Here's what we learned about building rational information-seeking agents... 🧵🔽

More detail including a code sample for trivially implementing linking in an agent on my blog: mbleigh.dev/posts/co... Links were powerful enough to build the entire web...I think they'll be critical for building the context for agents as well.

Discerning what is at the core, and how to communicate it with others, has taken years. This might be the closest I've come so far. Part 2 of The Interfaces With Which We Think is out now... Start at the intro: alexanderobenauer.com/think/ Start at Part 2: alexanderobenauer.com/think/2/

Recent developments like subagents and agent planning tools have got me thinking: what are the right abstractions for natural language programming? I wrote about how we're rediscovering structured programming but this time for prompts instead of code.

New post 🎉 Going back to my roots on writing about the inner workings of things, a breakdown of key-value databases and how you might make one from scratch: nan.fyi/database

An AI coding approach I've liked using, which I first used on this project: Vibe code the first version. Investigate the code for its shape, complexities, needs, failure points. Then scrap all that and engineer a new version.

Great analysis of spec-driven development tools and recommendations for how to make them more effective: more variety of workflows, smaller iteration cycles, and better spec review tools

Some thoughts on LLMs, Autopilot, and VibeCoding rahulpandita.me/blog/2025-10...

These new Windsurf codemaps look a lot like Project Copernicus from @kcieslak.io githubnext.com/projects/cop...

6/ That said, HRM and TRM are fascinating proof‑of‑concepts that show what’s possible with relatively small and efficient architectures. I'm still curious what the real‑world use case will look like. Maybe they could serve as reasoning or planning modules within a larger tool‑calling system.

We have 🤖 AI notetakers in meetings but continue to silo know-how every time we close terminals. Not just the how but also the why and what. Sign up for beta access to visr.sh: it's like as if Granola AI and #tmux had a kid.

Fun panel discussion on spec-based development with @kiro.dev team. Especially interested in upcoming features to ground the specs to verify that the work is done correctly

What, if anything, do LLMs change about interface design?

As autonomous agents tackle larger tasks over longer trajectories, we’ll need tools like Trailblazer to quickly walk us through how problems were solved so verification doesn’t take as long as solving it ourselve

Teaching LLMs to Plan: Logical Chain-of-Thought Instruction Tuning for Symbolic Planning arxiv.org/abs/2509.13351

My ideas are faster than my fingers so I've did some Figma prototyping and video editing to talk about some of my new ideas before they are actual coded reality youtu.be/2DRWUgK-RdA

Want to visualize the response format constraints on the LLM when working in a Jupyter notebook? Then you might be interested in my new project `litelines`. Litelines lets you visualize the selected path by the LLM. It supports a Pydantic schema as a response format, as well as regular expressions.

Sometimes you need to build the wrong thing to get in the right headspace to build the right thing

"Generative, Malleable, and Personal User Interfaces" by Haijun Xia (2025) www.youtube.com/watch?v=MbWg...

Advanced context continuity with Projects and the Claude Pruner bookmarklet whtwnd.com/austegard.co...

Need to rethink my gesture math and try some polarizing film, but basics of panning the canvas are in there.

Cursor limits me to 5 custom modes so I reserve those slots for coding-specific tasks and do all my broader design thinking in Claude Desktop instead. Interesting how small constraints in our tools quietly redirect our entire creative process.