Kyle Kelley

All people look like muppets yet some are more muppet like than others

Announcing nteract 2.0! www.nteract.io/blog/nteract... Native notebook app with managed python environments, realtime collaboration with agents, and some @deno.land to boot.

We got a DM the other day with a fun request from a local family: Could we find this nice BART train operator who always waved and beeped at this special little rider? And if we found him, could we deliver him a thank you card!?  So we got to work …

Silly agent. Asked for a happy little plot and it delivered.

Added ctrl-r search to nteract

Based on feedback from @rahuldave.bsky.social and @captainsafia.com I'm moving settings out to a proper window all it's own. nteract → settings

You ever just fully lose a discord server amongst all your discord servers? I cannot find @deno.land in my left bar so now I have to visit on the web to get "invited" again. Still no way to bubble it up. :(

I wish Dependents view in GitHub would let you filter out forks.

You can just research things. New from @jatucker.bsky.social & me at @brookings: Coding agents like Claude Code and Codex will likely accelerate research AND undermine institutional structures we built to support it.

me rn

I love when Claude says it's going to take 2 to 4 hours for something that I know will take it like less than 30 minutes in a code base.

Generative AI has the memory of a goldfish and a million eyes.

I need some way for my agents that operate on behalf of me on GitHub to indicate as such. Aside running and orchestrating these like ZERO places are making it possible to provide attribution and authentication for agents in a clean way.

Errors as product placement

upstream connect error or disconnect/reset before headers. retried and the latest reset reason: remote connection failure, transport failure reason: delayed connect error: Connection refuse

Speed of development wholly dependent on how fast testing and especially end to end testing can be done.

Wow. @adriansieber.bsky.social (re)implemented the Wolfram language in Rust github.com/ad-si/Woxi and exposed it as a jupyter kernel. That's crazy.

You would not believe the things I can now blame on AI.

Mmmmm scrolly terminal in @zed.dev's REPL output (@jupyter.org backed). PR: github.com/zed-industri...

Much respect to @anthropic.com for holding to standards. www.axios.com/2026/02/15/c...

What if notebooks let you install dependencies directly and when loading someone else's, they inform you of what's going to be set up in the environment? Yes this is a new desktop app.

I've been in the AI world for like two years now and I am continuously amazed by how many problems are no longer technical problems. The bigger cost of implementing automation is fine and institutional inertia.

Backlogs can only grow.

Damn GitHub is still having trouble. Maybe they should run some copilot on it.

Jupyter and IPython are NOT in the epstein files. This notebook, however, has seen things.

Ready the PR cannon. Branches standing by. Awaiting further instruction from @github.com

My theory here is the major labs have done very little or no RLHF on visual explanations and diagrams? Seems like a blind spot. Image generation alone or text generation alone are very well optimised but the combination seems uniquely hard for current models.

Not a day goes by that I'm not mad at Howard Jarvis and Paul Gann

Finally tried out PyO3 for python <> rust bindings with async/await support. Not sure what the best patterns are but it's way easier than I expected. github.com/alexjg/samod...

planting 40 million merkle search trees, in 20 years we'll have a whole lot of merkle search forests we can log the merkle search forests for merkle search timber which we will then use to construct a merkle search floating city

I need cat-aware undo