with all the deskilling and gell-manning (etc) we're gonna get from automated thinking machines, now is the time for the 'tools for thought' crowd to make tools that help you think things through for yourself.
tools to do *more* hard thinking.
it should feel effortful. cherish that feeling.
Also, when it says: I would recommend the first approach so you don’t have to refactor as much code.😂 I have some bad news for you buddy...
You’d be surprised at how human-like they drive now. But yeah they definitely act weird when they’re OOD
Congratulations Tom! Claude Code is a fantastic product
Values-as-Spec is the process of specifying what values the product is committed to protecting; how the product involves the user in its processes; and what the product explicitly will not automate or replace. What it does, won’t do, and can’t do.
@seanx.bsky.social
dearhermes.com/read/kfniw9y...
Self-modifying AI agents: "There are no configuration files to learn. Just tell Claude Code what you want. Or run /customize for guided changes. The codebase is small enough that Claude can safely modify it."
github.com/qwibitai/nan...
If you're a library author, it's a pain to figure publish agent skills in a way that your users will actually see.
If you're a developer using libs, it's a pain to make sure you're using the best and latest agent guidance for them.
Let's solve it! I'm proposing .agents.lib
That seems to be who Claude Cowork is designed for
Count me in!
The Sony Bravia XR chips do AI scene recognition as part of the image processing
📝 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
Big design up front sacrifices agility. Vibe-coding eventually hits a wall. Finding a way to control entropy while prioritizing exploration and innovation is the challenge. We're redefining what it means to be in a state of flow.
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
Thanks for sharing! Claude Teams does let you DM individual agents through the CLI so this future is very near
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
I was thinking of making this more systematic and having the agent automatically create blocking subtasks if a task fails verification but I need to think through the design in more depth
Thanks Brian! There's no specific flow for when the agent gets stuck. The user can always spin up more Claude sessions to satisfy any leftover acceptance criteria. And then re-verify...
You can install the extension from atelier.dev I wrote some docs to explain the install process. Let me know if you have any questions about it
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.
Early beta is available to install at atelier.dev. Would love to hear what works and what doesn't
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
bsky.app/profile/narp...
I’ve been doing this too. At first I thought that I had screwed up a vibe coded the wrong thing without pausing to spec it out properly but I think there are some solutions which are only obvious when you’re inside the problem looking out
I agree. A commit is a natural place to ground user intent in a specific snapshot of the code. Maybe there's a way to link them together? Like "here's the code that resolves the issue" but also "here's the series of decisions which led us there"?
