Paul Taylor

Unpopular opinion: the best AI infrastructure tools won't come from the big labs. They'll come from open source. 🛠️ Labs never see the weird failures, the vendor-specific ones, the ones that only happen at your company. A community does. The labs have the models. The community has the failures. The failures are the moat. 🏰

We recently ran a 7-day hackathon. 🛠️ → A pipeline rewrite shipped at 2:30am. → 10 PRs merged in a day. 23 total. → +36pts on the hardest Kubernetes failure category. +400 GitHub ⭐ in 7 days. This is what building in public looks like. See you at the next one! 🚀 github.com/Tracer-Cloud/o…

We benchmarked OpenSRE against 452 broken Kubernetes clusters. Same GPT-4o model. Same tools. Only difference: the investigation loop. 🔍 Hardest fault type in the benchmark: OpenSRE scored 56% vs GPT-4o at 20%. That's what thinking harder looks like. 🚀 github.com/Tracer-Cloud/o…

Big shoutout to Greptile, our newest sponsor 🦎 Less guesswork navigating a codebase, more time actually fixing things, that's the instinct behind their tool, and it's one we share. Proud to have them in our corner as we keep building OpenSRE fully in the open. → github.com/Tracer-Cloud/o… → greptile.com

big two weeks for opensre ↓ • GitHub Actions - investigate CI/CD failures directly from your pipeline • xAI Grok added as an LLM provider • 6 new integrations: Azure SQL, MongoDB Atlas, Snowflake, Supabase, OpenObserve, Trello • Hermes eval suite - measure RCA evidence quality, not vibes • CLI: session resume, cleaner commands, silent errors fixed github.com/Tracer-Cloud/o…

🌞This is big Local AI news! A new open-source Computer-Use LLM has just launched. Holo 3.1 is H Company’s (🇫🇷) new local computer-use agent model that beats Qwen3.5-397B, Kimi-K2.5, and Sonnet 4.6! Since it is built for local deployment → ⬩ Runs fully on your machine (MacBook, Windows PC, DGX Spark, RTX Spark) ⬩ Based on Qwen architecture, specialized for GUI understanding & computer control ⬩ Optimized checkpoints: NVFP4, FP8 & Q4 GGUF (0.8B to 35B sizes) ⬩ Strong gains: 79.3% on AndroidWorld benchmark (35B model) 💻 Comparison to Qwen3.5: Holo 3.1 is fine-tuned specifically for computer-use agents (screen understanding, planning, clicking, navigation). Better at real GUI tasks than general-purpose Qwen3.5, especially when running locally.⚡

Workflows are the biggest upgrade to Claude Code’s capabilities since skills and subagents. I dove deep into it with @sidbid to figure out best practices, examples and more.  I’m particularly excited about the non-technical tasks it enables for Claude Code.

Following the great first OpenSRE community call on Discord, we’re hosting the next one tomorrow, Thursday April 16 at 17:00 UK time. discord.gg/A9J2dHdH?event… Come join if you want to jam on ideas, debugging, open-source, and building. See you there!

OpenSRE just hit 5,000 stars on GitHub! ⭐ What started as an idea is now a community of thousands building, contributing, and shipping together. Thank you all! Lots more to come. 🚀 github.com/Tracer-Cloud/o…