Last weekend my mom made an app to run registrations for a half marathon in my hometown in Italy. She and a few local high school volunteers checked in 2000 runners. Signups, bib assignment, race packet pickup, payment, all of it. To build it she used a tool I made called Spunto. It’s a “Claude Code for my mom”. You chat with it and it gives you a link to a web app ready to use or share. My goal with this project is to share the joy I feel building with coding agents with friends and family who haven’t seen this glimpse of the future yet. Seeing my mom have that “aha” moment was exactly it. My cousin used Spunto to plan a trip to Thailand, his roommate made an app to split apartment costs, my wife built a tracker for our 5mo boy’s feeding schedule that we now use every day, and more nice stories like this. This was inspired by how @steipete helped people beyond software engineers “see” what agents are with OpenClaw. Amazing times to be alive. We should try harder to bring more people along this journey. It feels great when that happens. If you are curious: spunto.dev/

A logician lends a formal system twenty euros. A week later, he asks for it back. "I can't prove I owe you anything," says the system.

Plainsolid is getting better. It's a CAD tool to replace SolidWorks at my startup. It also has CLI, MCP & AI suggestions with context from UI and hints from user. Saves us 10k/year. We add features when we see something missing during real work. github.com/oloapiu/plainsoli…

I used Fable and Astra to code it. Impressive what can be done with very little oversight now. It is far from being as feature complete as SolidWorks, but for my workflow it is close to enough. We mostly design weatherproof enclosures with PCBs and sensors inside and antennas, solar panels, connectors, vents and brackets outside. Our ME designs are more like concepts, the real DFM is done by the CM with their own tools.

@antirez did you read it? I think you would love it. Funny enough, Sol recommended it to me after interviewing me on my sci-fi tastes

The gist of it is humans getting in touch with aliens of a very different physiology that live on a neutron star. One of our seconds is equivalent 12 of their days. At first they are barbarians and us gods teaching them about science society and art. By the end they leave us in the dust, but with gratitude.

Am I wrong or is buzz is a great way for many teammates to share a single Claude Max sub?

I think buzz hits the spot for human/agent collaboration at work. It is what we've been trying to hack together on our Slack in the last months. It's the right surface. But when I showed it to our devs the first comment was "I miss the kanban view, I'd rather use multica". I think that misses how the rest of the company works: non-devs ask for help in chat, and moving those requests into Trello by hand is painful and looses context. But I agree that the bird-eye view where you can prioritize and assign is useful. So I experimented with buzzboard: a kanban over buzz. CSM asks an agent in chat to file a ticket → it lands on the board → devs triage it and assign it to an agent → the agent works the card and reports in a thread. The board is fully stateless. every card is a signed Nostr event on your community's relay. No server, no database, teammates run their own copy, same board. I think this matches the spirit of buzz nicely. demo: oloapiu.github.io/buzzboard/… or: npx buzzboard code: github.com/oloapiu/buzzboard (help me make it better). @jack this is Nostr as a work OS. Curious what you think.

Interesting. Testing on my new DGX Spark. DeepSeek v4 Flash q2 on DS4 worse than Qwen3.6 35B q8 on this benchmark: github.com/SeraphimSerapis/t…

Reproduced @MiaAI_lab's tool-eval-bench result for Qwen3.6-35B-A3B (Q8 GGUF + llama.cpp + MTP) on our own DGX Spark: 90.9 vs their published 91.0. Same benchmark, 8 trials, seed 42. That's a reproducible recipe. 👏 Bonus head-to-head on the same machine vs our tuned vLLM stack: nightly vLLM, all-NVFP4 4-bit checkpoint (unsloth), Marlin MoE, FP8 KV cache, MTP k=3, plus the kernel workarounds sm_121 still needs. The 8-bit llama.cpp setup wins on quality, way wins on run-to-run consistency, and ...surprise... is even faster than 4-bit vLLM. Its only loss is memory (39 vs 22 GB). Table below. Recipe: github.com/MiaAI-Lab/Qwen3.6…

DGX Spark arrived yesterday. Had a fun time setting up vLLM and serving Qwen3.6 35B A3B NVFP4 for a custom agent I've been working on. I'm new at this so it took a while. Learned a lot. Sharing my journey here in case it is useful to someone. The main tip: have claude code or codex drive lol xitter.cf/oloapiu/status/2073437…