Software Architect | AI | .net | DevOps | k8s | Founder https://xitter.cf/t.co/sFRSsEofCX
London
Joined August 2026
- Tweets424
- Following1.2K
- Followers555
- Likes1.9K
Pinned Tweet
I gave three AI agents a maze, notebooks and radios to see if they could help each other escape.
Some shared directions. One walked into the same wall 71 times.
A small experiment in getting lost together.
Spencer retweeted
meet buildnotch. a minimalist windows notch made for developers, founders, designers, power users.
github stats, favorite AI apps + usage, personal projects, repos, and 7 amazing focus timer styles.
might be your last chance to get it: buildnotch.app/
Can one browser click stay one trace through FastAPI, Pydantic AI, MCP, a .NET agent, SQL and HTTP?
I wired it up with OpenTelemetry, sent it all to @Pydantic Logfire and checked every parent span. It held, after one wrong turn worth sharing.
Spencer retweeted
Sat on 968 followers right now, exactly 32 away from 1,000.
I set a goal to hit 1k last Sunday so I am running a bit behind schedule, but I reckon we can get it done today.
Any indie hackers and builders fancy connecting?
Who is pushing me over the line?
With default settings, Qwen Code's telemetry included my prompt, the files it read, and memories from other sessions. Codex exported file contents in tool output. Claude Code put my email on every metric.
What three coding agents send over OTel:
Part 2 of my coding-agent OpenTelemetry series: what Jaeger and @Pydantic Logfire made of Claude Code, Codex and Qwen Code and Deepseek Harness, and how LangSmith, Langfuse and Braintrust compare.
Part 2 of my coding-agent OpenTelemetry series: what Jaeger and @Pydantic Logfire made of Claude Code, Codex and Qwen Code and Deepseek Harness, and how LangSmith, Langfuse and Braintrust compare.
got my 2nd ever hackathon tonight. a short one again. need to think more outside the box on this one. got a load of lovable credits left over from last weeks hack, so i can probably put them to good use.
Spencer retweeted
Meta is about to give away $1,000,000 to AI builders.
Most people will find out after it's already full.
The Meta Global AI Developer Hackathon:
• $1M cash prize pool
• Run by Meta Superintelligence Labs
• 10 days, fully virtual, open worldwide
• Every participant gets $150 in @Meta Model API credits, even if you don't win
• You build on Meta's newest models
Applications aren't open yet. That's the point.
Join the notify list now so you're in on day one not day five.
Bookmark this. Tag the friend you'd build with.
Link below 👇
dev.meta.ai/events/global-ha…
Added a 4th agent since writing this: DeepSeek Harness with LoongSuite's OTel plugin. Cleanest trace of the four: standard gen_ai.* names, no prompt or code exported by default, and it lights up Logfire's Agents view.
Spencer retweeted
👉 Introducing Gimme Gimme Ads 1.1.
This is for the people working in paid ads.
Stop paying thousands in agency fees. Stop getting AI slop for ads.
Just use Gimme Gimme and get insane ad creative for your brand in seconds.
> Paste your product or website URL
> Your brand, products and tone of voice are researched
> Winning meta ad formats in your industry are learnt
> You receive insane ads
Try it on your store now: getgimmegimme.com
P.S. I may have been too excited in this demo, my bad.
LangChain told me three times that my model didn't exist. The request on the wire was POST /chat/completions, a 404, with no /v1 in the path.
Vessel is the local proxy I built to show me that layer: what my agents actually send. How it works:
anyone here going to the openai london hackathon tomorrow? will be my first hackathon - bit nervous now lol
Nice to see jev-type model support added natively to Ollama. People seem to be converging on the same API request/response format for it which is good news. I doubt OpenAI's upcoming Decision api will stick to it though.
