Devin Stein

@RhysSullivan @claudeai will try this out

@devstein64 @claudeai I liked the JS slides framework for making slides with Claude, was really easy to use

@claudeai design for generating slide decks is amazing. claude design for editing slide decks is one of the most painful UX experiences I've had in recent memory.

@RhysSullivan hadn’t thought about this. I definitely don’t care nearly as much as I used to

do you still care about backwards compatibility / breaking changes in libraries now that agents are writing most of your code?

@steipete @sveltejs would be very happy to see a @sveltejs comeback

Been using @sveltejs for a few projects lately, it's quite a nice alternative to React, fewer gotchas and complexity and Codex handles it really well. svelte.dev

@sarahwooders @Letta_AI but how do you name them?

Custom agents let you scope responsibilities in ways that make sense for your org (like with real employees) At @Letta_AI we have 4 AI employees in Slack with overlapping tools, but different responsibilities: - Ezra: Customer support + issue creation - Overload: Manage issue->PR pipeline (including implementation) - Noam: Manage training/eval jobs - Brad: Office management + ops

been saying this for years!

@benhylak @raindrop_ai loved the onboarding experience @benhylak

431 stars in 24 hours. to put that in perspective: langfuse took 2 months to hit 1,000. thank you to everyone who tried @raindrop_ai workshop.

tend to your knowledge garden

Untended docs rust, and spots spread until your team's momentum slows. How fresh are our docs right now? Wouldn't you love a solution that lets you track this over time? Build a solution with us and start scoring documentation freshness with each incoming PR. go.dosu.dev/yTM4gkT

Agents use systems like databases, search engines, and observability platforms in a fundamentally different way than humans do. They tend to do a ton of small work in parallel. They are far more bursty. They are a much smarter client than humans are, altering how they write queries or how they want to use an engine. They prefer to rapidly iterate as they try things, failing frequently but correcting in response. One of the biggest questions many systems companies are thus wrestling with is - how do we redesign our system for agents? I wrote what I believe is one of the most comprehensive, detailed overviews of the patterns with which agents tend to use systems, and the architectural elements this increases the importance of. I touch on quite a bit of research in this category, as well as many examples of interesting startups at the forefront of thinking about this topic - like @Bauplan_labs @turbopuffer @TigrisData @mesa_dot_dev @p0 @daytonaio @neondatabase @usefiretiger @FireboltHQ open.substack.com/pub/davistreyb…

Exceptionally excited to have @auchenberg join our team in NYC to double down on our work in software infrastructure. He and I are both convinced that this will be the most exciting time period to invest in infra ever. Every SaaS app is becoming headless infrastructure for agents. All existing infra systems are being redesigned for agents as their users. Agents represent entirely new distribution rails for developer tools ("Agent Experience"), giving a new GTM wedge for infra startups to build against. And every agent is becoming a coding agent, so the market for developer tools is now *all applications*. To mark him joining, we wrote up an overview of why the combination of these trends means the infrastructure market will likely 1000x, if not more, over the next decade. innovationendeavors.com/insights/a-dec…

friend

@zeeg imagine

imagine if LLM vendors focused on being infra instead of trying to build (mediocre) versions of all of their customers products

an austrian coworker once told me they much prefer doing work in english because the sentences are ~30% shorter. makes me think about token efficiency of languages for LLMs. which language is the most token efficient?

@sarahwooders mesa.dev

Is anyone building github for agents? Specifically: - API/SDK interface - make millions of repos no problem - fast

@mitchellh slop is a great personal tool. I love it for experimenting with different designs. it breaks down for collaboration, reviewing slop from other people feels burdensome.

AI slop is good, actually. Slop is what enables fast parallel experimentation. The etiquette and skill is understanding the boundaries of where slop exists and the extent to which it should be cleaned up and how. A few examples: I’m working on the internals of some system right now. The API and GUI of this thing is fully zero shame slop. It’s horrible. But it lets me focus on the core quality while shipping a usable piece of alpha quality software to testers (transparent about the slop frontend). Similarly, this system has plugins. We sent agents in Ralph loops overnight to generate dozens of plugins. The plugins are slop. The quality is bad. The plugin API/SDK is absolutely not done. But we can test a full GUI with a full plugin ecosystem. When we change the API, we can regenerate them all. The cost of change is just tokens, the velocity is incomparable to before. I built Terraform. We tested and shipped TF 0.1 with about 3 very weak providers. Because we ran out of time. Building was slow. And when we changed our SDK the cost was immense. Totally different today, 10 years later. Today, I would’ve slop generated 100 providers (again, with transparency and cleanup later, but just to prove it out). As an anti example, I would not PR this (without prior warning) to another project. I would not throw this onto customers without full review or transparency (as I’m already doing). I would not accept first pass slop. It’s almost never right. Slop is a tool. And like anything else it’s not blanket bad or good. The context is everything.

@zeeg @yehiaabdelm doom scroll

@yehiaabdelm what would you do with it

We need a sentry mobile app @zeeg

knowledge infra

@gs_ai_ Can’t get past the gloves

We are back. After one year of quiet building. Introducing GENE-26.5, our first robotic brain that takes a major step toward human-level capability. For years, robotics has struggled to learn from the world’s largest and valuable data source: Humans. Solving it means rethinking the whole stack from the ground up: - A robotics-native foundation model. - A 1:1 human-like robotic hand. - A noninvasive data collection glove for motion, force, and touch. - A simulator that turns weeks of experiments into minutes. GENE-26.5 is trained across language, vision, proprioception, tactile, and action. We designed a set of tasks to test how far we can go with this new paradigm. Fully autonomous, 1x speed, one model, same weights. (Enjoy with sound on) We are approaching the endgame for robotics. And this is just a beginning.

@claudeai design is clearly vibe-coded, but still very useful for product UX exploration