Abhij — oss/acc retweeted
Sooo excited to welcome four new projects to the Agentic AI Foundation! 🎉
- Agent Skills
- Open Responses
- OpenEnv
- OpenSandbox
Huge thank you and congrats to Anthropic, OpenAI, Hugging Face, and Alibaba for contributing these projects to a neutral home ❤️
Truly honored to help support the growth of each.
Abhij — oss/acc retweeted
Spiral’s @moneyball is speaking today at AGNTCon + MCPcon in San Jose about @goose_oss’s evolution into an open-source agentic SDK.
10:35AM in room 230C
Abhij — oss/acc retweeted
One of the best ideas @spiral_xyz has had
Project Loupe - AI powered security scanning for Bitcoin OSS Projects xitter.cf/i/broadcasts/1nxnRBMEo…
Block has joined the Open Secure AI Alliance, hosted by the @linuxfoundation. AI security gets stronger when people can build, test, and share defenses in the open: block.xyz/inside/block-joins…
Abhij — oss/acc retweeted
Friday lightning talks are back at Bitspace, and #Bitaloo (@shourya742) is the because. They’d stopped for a while, and then he started them up again and somehow got the rest of us to plan our Fridays around them.
Two weeks ago it was deterministic simulation testing. Last week was coherence and the orphan rules in Rust. He’s been deep in AI for quiet a while now, and basically everything he figures out ends up in front of us, which is one of the best deals any of us has ever gotten.
We call them lightning talks, but each one #Bitaloo did, ran more than an hour, so it’s a bit less lightning and a bit more thunderous, and all Bitspace regulars have mutually agreed to call his talks "Thunder Talks" from now on.
It’s been delightful and fun to see this happen. Someone gets curious about something, there’s a room full of people happy to listen, and a couple of hours later a bunch of us understand things we had no idea about the week before. Having a place where this can happen casually is a big part of why we’re here at Bitspace.
The next AI Thunder Talk is tomorrow at 8pm. The start time is the only part we’re sure of, which is why dinner will be getting delivered to Bitspace!
Abhij — oss/acc retweeted
Open Source AI: It's Time to Hit the Gas
Last month, we brought the community together at the Open Source AI Summit in San Francisco's Presidio. Here are our top 5 actionable takeaways. 🧵
Abhij — oss/acc retweeted
New vertical: AI Caramba lets the goose team open up about anything AI. In this first edition, @dosinga explains how goose replaced the agent loop with a state machine that can be turned off anywhere and resumed like nothing happened. Where does it go in between? Hell if we know.
Abhij — oss/acc retweeted
OpenSwap team has been constantly pushing boundaries to enable a truly decentralized atomic swap market over Bitcoin & LN.
With the recent CPOF dependency on central swap servers, OpenSwap is the only solution ahead to decentralize the swap market.
Important stuff brews in silence.
Abhij — oss/acc retweeted
we got agent swag y'all! 🤖
New: AAIF merch. Ship agents, also ship t-shirts. 🧢👕
merch.aaif.io
Abhij — oss/acc retweeted
43 open-source bitcoin repos reviewed and zero LGTMs.
agree @moneyball
Shared context sets the pace.
On The AI report, Block's Head of AI Capabilities @bradleyaxen talks AI memory, the thinking behind Buzz, and why the future of work is multiplayer.
Watch here: youtube.com/watch?v=a3662l7b…
An open source team moved its day-to-day development into Buzz. Across matched workweeks, MeshLLM merged 56% more PRs and cut median merge time from about 7 hours to 4. More on the setup, data, and caveats from our engineering team: engineering.block.xyz/blog/s…
Abhij — oss/acc retweeted
Angie Jones has officially entered her protocol girly era. 💅🤖
Her take on WebMCP: what if tool calling lived directly in the web page itself?
That gets especially interesting when you and your agent need to work together in the browser but for WebMCP to really take off, developers have to start building for it.
Hear more from Angie and other builders pushing Agentic AI forward at AGNTCon + MCPCon NA: bit.ly/4vkntyD
📍 San Jose
📅 Oct. 22–23
Abhij — oss/acc retweeted
Context can be something that evolves continuously
For example you can look at a large tool result (grep etc) and ask what parts of the output belong to stay
You can route between models (like the previous demo I posted) with a slimmer set of more relevant info
Abhij — oss/acc retweeted
A useful provocation
“how would you design a coding agent if LLMs had no KV cache?”
A lot of agent architectures rely on a conversation that grows until compaction without much consideration what’s in that conversation
System one models can help with new approaches
sharing some notes on typesafe 🤝 coding agents:
docs.google.com/document/d/1…
we likely will never have time (ever again) to play ourselves, but hope the that the community goes WILD (and makes me look like a naive idiot)






























