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.

Is there a name for the pattern where you try building the same thing (feature, project, ...) a bunch of times with different agents to get a sense of the implementation design space? I usually run things at least 2 or 3 times until I "feel" what is the way I want it to be built.

Software engineering is evolving into software research. AI can now recreate most known solutions, so human value shifts from "how to build it" to "what should we build next?" The future belongs to those who discover new problems worth solving

Love this analogy from the RepoAudit paper. LLMs with huge context windows are like putting all your code on a single PowerPoint slide and then projecting it onto a 1000-foot screen arxiv.org/abs/2501.18160

Don did a good job explaining this here dsyme.net/2025/08/27/o...

I've been thinking about hypermedia and hyperlinks might evolve with LLMs. aifoc.us/hypermedia/ It feels like there is a lot of opportunity how links work and the value they can provide people.

A bit more about the creative word search tools in Phraselette: The 'Reader' well lets you to mock a panel of readers to examine specific word choice.

Skills are the new currency: https://blog.davidhoang.com/2025/08/17/skills-are-the-new-currency/

working with AI makes me wish more than ever that we had first-class layered undo. git is just 1 layer, yall. editor undo is a layer on top. i want several more layers, eg 1 to scrub the chat and codebase in tandem. 1 to cherry-pick specific changes made by the agent. don't tell me to multi-repo!

This new prompt optimizer from OpenAI is an exciting development. You can give it a prompt that you've already written and it will suggest improvements to your prompt and explain its reasoning for the changes platform.openai.com/chat/edit?op...

"Making these tools available to the systems we build, and specifically to the AI agents we build, gives them more power and leverage. Not by encoding the way we think, but by encoding the things computer science has learned about the way the universe works."

Can LLMs reason by analogy like humans? We investigate this question in a new paper published in the Journal of Memory and Language (link below). This was a long-running but very rewarding project. Here are a few thoughts on our methodology and main findings. 1/9

Chat is a great modality for starting open-ended single-player explorations. Chat is a terrible modality for managing long-running, structured, or multi-player tasks.

Great insights into writing custom rules that work. "The effective approach is to make the AI itself a partner in refining them." "Analyze our recent conversation. Based on the guidance and corrections I provided, suggest improvements to my .vscode/copilot-instructions.md file."

I wrote up some thoughts about why we’re seeing a gap in AI tool use among different types of developers narphorium.com/blog/top-dow... 1/4

I've been trying to articulate why the fawning, complimentary responses from AI chatbots feel so insidious to me. I've finally figured out how to explain it. Wrote a long piece on how current model training and design choices threaten our critical thinking skills: maggieappleton.com/ai-enlighten...

Reality Proxy: Fluid Interactions with Real-World Objects in MR via Abstract Representations arxiv.org/abs/2507.17248

I do this too. After an AI agent writes a lot of code I start a chat with my refactoring prompt to make sure it’s using the right abstractions open.substack.com/pub/worksonm...

Transclusion for context engineering www.fuzzycomputer.com/posts/spectree

As a big Jupyter notebooks user I can’t wait for better AI agents experience like this

Excited to pore over this new book from Buddy Bindery. Thank you @alexanderobenauer.com for sharing your artistry with the world

Turning "drinking water" into an achievement that you brag about has to be one of the most millennial things ever

Really neat idea for combining the strengths of different models

Great idea! This is something which senior engineers often do in their heads but I’ve never seen tools visualize it like this

Natural Language Outlines for Code: Literate Programming in the LLM Era arxiv.org/abs/2408.04820

AI codegen is walking us backwards into literate programming. Instead of jumping directly to code, I now dump my half-baked ideas into md files in a "specs" folder and let the agent rip. Or write a comment describing the function I want and AI completes it. My code has better docs and tests.

okay, vague idea: integrating LLMs + block based programming overcomes the biggest limitation of block editors (looping control flow etc.) internally, generate python jupyter-like snippets, but user should only ever see outputs. Draw edges to describe dataflow (demo whipped up using @tldraw.com)

This was a great talk which has nerd sniped me into building a data labeling widget with @anywidget.dev

I am PUMPED to finally share what we’ve been working on: 🖥️ Introducing the Gemini CLI! It can code, sure, but with access to your system, files & MCP servers, it can: 👩‍🔬 Do research 💽 Organise your MP3s 🪾 Resolve rebases 🔬 Even strace that weird hung process Check it out: goo.gle/gemini-cli

Wigglystuff has a new widget, thanks to Trevor Mantz! You can click/drag/add/remove/change items in a list visually. Should be useful for some bespoke annotation tasks!

Never a better time to build personal tools using agentic coding techniques. I'm working on a Trello replacement with integrated tool-using AI chat sidebar. All of the features I want, none that I don't.

Double Diamond design is moving inside the traditional programming development loop. You're not just iterating on the code but also on the idea.

This is a smart use of UI context. When users are coding in a Jupyter notebook they're probably in a more experimental mindset so unprompted AI feedback is welcomed.

We need better tools for expressing the intent of software and not just the implementation details

This is a great idea. AI tools should be away of my design system and be able to create new features by reusing and extending existing components

New longform @inkandswitch.com essay! 📜 Malleable Software: Restoring user agency in a world of locked-down apps by me, @joshuahhh.com, @pvh.ca and @seaofclouds.com inkandswitch.com/essay/mallea... It's about why people need agency over their software tools, and how to make that happen.

The "liquid" part of Liquid Glass feels more interesting than the refraction effects. The liquid transitions on the dynamic island were a stroke of genius and I'm looking forward to more of that across all my devices

New Post: MCP Resources Are For Caching This is a quick tour of what MCP resources actually are. And more to the point, what MCP is supposed to do (and not supposed to do). timkellogg.me/blog/2025/06...

I use playwright-mcp to drive the browser, but why not go directly to CDP to get full access to networking and performance as well as DOM etc? Stefan Li wired this up with devtools/mcp! Love it! autoconfig.io/devtools-mcp...

it's called inkwell. an mmo that dreams up new worlds & adapts.

Introducing The Darwin Gödel Machine sakana.ai/dgm The Darwin Gödel Machine is a self-improving agent that can modify its own code. Inspired by evolution, we maintain an expanding lineage of agent variants, allowing for open-ended exploration of the vast design space of such self-improving agents.

“We no longer have to think like a computer. We can just speak like a human. This obliterates the idea of an interface as a translation layer between humans and computers. Interfaces aren’t totally going away, but the best ones now get out of the way and just let us be human.”

Claude 4 vs Claude 3.7 is like when you're in a meeting and you can tell who read the doc beforehand vs who's skimming it live just based on the questions they ask

“When a senior engineer joins a team and opens a large monorepo, they don’t read isolated code snippets. They scan folder structure, explore files, look at imports, read more files. That’s the mental model your coding agents need to replicate…” open.substack.com/pub/pashpash...