friggin' love Conductor.build

cursor composer zoom zoom 🏎️

Nice office and even nicer guy! Recommend 🅰️➕➕➕

oh right, it's that one weird week where the US and Europe diverge on daylight savings time

90% of the time when an LLM fails to do a thing for you, it's a lack of proper context, not model capability.

rooms are where life happens. we create, workshop, collab, party, veg out, bring up our kids... in rooms and we do it together why would I want a computer strapped to my face? instead: a computer we can walk into... and walk out of I am all-in on augmented environments.

Why does Claude Code operating directly on files in my Obsidian vault feel so much better / different than Claude Desktop + an Obsidian MCP server?

I finally did the thing: pointed Claude Code at my Obsidian Vault's markdown files. Oh. My. Goodness. This is great!

I think this succinctly explains the rapid adoption of LLMs in a variety of domains. There's something tipping point for accuracy where the generality wins become more important than 100% deterministic predictability.

I'm finding it's more valuable lately to spend my time on "build the software factory" than to actually build the software directly. Improving the quality of work and parallelization across my AI agents is a huge force multiplier.

🎥 Tomorrow 10am ET/7am PT Continuing to evaluate using the Claude Agent SDK with @livestore.dev on a livestream. Come hang out! youtube.com/live/bOVeT-s...

🎥 Starting now! I'm livestreaming building a LiveStore app that uses the Claude Agents SDK as the "agentic loop" and interacting with the LiveStore app through custom tools. Should be fun! Come hang out. www.youtube.com/live/5-1TF51...

Probably gonna livestream building a LiveStore app with the Claude Agents SDK.

LiveStore Hands-on #1: Work Squared Friday, Sept 26 @ 5pm CEST / 11am ET (tomorrow) Watch @schickling and @jessmartin hands-on with a real LiveStore app: ▪︎ Architecture review ▪︎ LiveStore best practices ▪︎ Live debugging ▪︎︎ Source available luma.com/4s7qev78

Book-to-soundtrack experiment - a projector setup where I send the book image off to an LLM for recommendations and then play with Spotify. Meant to be kind of like putting on a record but for objects. (I clipped out the loading times.)

I'm demoing worksquared.ai and how I'm using livestore.dev to build it in about 40 minutes. Sync engines are awesome! luma.com/dp4j8a3e

Bless you, Conductor

The definitive UI paradigm for AI has already been established. It's Chat. It's unfortunate. But thems the Facts. Just like you can't reinvent the steering wheel when designing a car, you can't escape Chat.

@danversfleury.bsky.social and I are walking this path of pain together and trying to build our way out of it. It's so... fun?

I've made a *lot* of @tldraw.com's over the years, but these two might be my most favorite, ever #nocontextwhiteboards

At @sociotechnica.org we've been working on an "agentic environment" called Work Squared for a few months and it's starting to feel truly useful! youtu.be/1yfpGT6B50w Core concepts: - Agents and humans can collaborate on all of the work products - Make it visual - see everything that's happening

A horror story, in 6 words: Your branch and 'main' have diverged.

AI advantages existing platforms. The more data/tools you have, the more AI can do. Incumbents will sherlock every innovation. Yet AI also requires rethinking software from scratch. What will prevail: platform dominance vs startup disruption? notes.jessmart.in/Lab+Notebook...

I'm running the same experiment: using our own AI tool to help me get my life in order. Excited to see where this goes.

@danversfleury.bsky.social you should check this out - you'd like it

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...

Been thinking a lot about "building my software factory." I have a pretty solid process and can keep up with two Claudes running in parallel, but I need to build some specialized agents: - code review (just first pass, I still review) - design work (claude can't match a visual spec yet)

In the coming world of infinite software, we'll take software for granted. It'll be an after thought. Of COURSE the features you want will be present at the time that you need them. To unlock the potential of infinite software, we need a new substrate.

Given how much the process of software production is changing, the rational thing to do is budget 20% of my time every week to simply capturing those process improvements. Build the machine that builds the machine.

he's right!

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.

Friends: make beautiful things.

I miss liner notes

Just read through the CLAUDE.md doc for a project I'm working on and made a bunch of updates. Some of it was just plain WRONG. 😳 Previously my view was "CLAUDE.md is for Claude. Let Claude keep it up to date." I think now I'll review changes to make sure false info doesn't sneak in and linger.

Here's the `/docs` directory for a project I'm coding with Claude Code really helps having all of this context inside the project

Wrote this surprising sentence in my morning retro: - Need to book "learning time" the way I have to book "exercise time". With the machine doing more of the implementation work, have to be intentional about learning how things actually work.

In this Cause & Effect episode, Johannes Schickling talks with David Golightly about how MasterClass built Cortex with Effect. Cortex is the voice AI orchestration layer powering personalized conversations with celebrity instructors like Gordon Ramsay and Mark Cuban. www.youtube.com/watch?v=x2bU...

iykyk

oh no, I need to not look at this... this is one of my pet problems I want to solve but don't have the capacity to pick up right now. focus, focus. 🧘

Took a few minutes to demonstrate how I’m working with AI for software engineering lately: youtu.be/MPbRGZtIHu4

“Elicit can extract technically complex variables from an enormous volume of papers with researcher-level accuracy – that’s incredible.” Read how Formation Bio, an AI-native pharmaceutical company, uses Elicit to to power hyper-efficient drug development: blog.elicit.com/case-study-...

I'm not a Fancy YouTuber™ 🤓, but I've done a lot of livestream coding on that site over the years (100+ hours!). Finally got around to updating my "channel front page" If you're into live-coding, you might like some of it. www.youtube.com/jessamartin

let it flow!

Time to go for a walk outside and talk out loud about the next feature I'm building. 🚶‍♂️‍➡️ Forgetting all the other LLMs, whisper alone has changed my life.

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.

background async SWEs is such a crazy unlock: 1. point a few claude codes at different github issues 2. go do yard work 3. come back to review PRs, test the auto-generated previews 4. make a few comments 5. go to the store to get dinner 6. come back, test again, merge 🤯

Finally got to follow Kay's advice from years ago to look into Jerome Bruner more deeply... Danvers and I are working on embodying these ideas in a system and it's really helpful!

Cloudflare gets it

this is the feature I'm most curious about in the upcoming Apple releases

Perhaps BlueSky needs to differentiate along a different access than “content”. Different forms of media, enabled by atproto. Twitter got to be Twitter in large part because of an interesting and open API.