For coding agents, "skills" are a great way to automate repetitive workflows, but how can we tell if they're working at scale? We did a deep dive on how you can log, monitor, and improve agent skills, with a real example of building a customized PR review skill.

We loved chatting w/ @ashimmy on #TechstrongTV 📺 Our CEO & Co-Founder @rbren_dev sat down to discuss the importance of #opensource, where we draw the line in terms of monetization, and the overall vision for the OpenHands platform & community ⭐ Full interview here: techstrong.tv/videos/intervi…

Claude Fable 5 is out and we're working on testing it on the OpenHands Index! Will post results when we have them.

Agreed. My basic agent loop is this: 1. Organize my to do list: pull in context from email, slack, github, and linear and add linear issues for any missing things 2. Walk through the to do list in order of priority, tell me the current status, ask what to do 3. I delegate a sub-agent (using the OpenHands cloud api typically) to work on that task 4. Once I have many tasks going at once, check on the ones that finished in order, and take any action I tried a more automated version of this where I skipped step "2", but eventually it resulted in a lot of slop, so I think the human decision step is still pretty useful.

Thanks James for recognizing @OpenHandsDev performance (And we do have a OpenHands Cloud where lots of folks get started very quickly - app.all-hands.dev/)

Want to see what "AI Agent Orchestration" looks like in production? Don't look at demos. Look at these open-source projects: 🛠️ 1. OpenHands 🔗 github.com/OpenHands/OpenHan… Why it matters: • Autonomous software engineering agents • Isolated sandboxed environments • Secure code execution • Human-in-the-loop reviews • Web UI for supervising agents OpenHands is one of the most mature open-source implementations of AI-driven software development, designed around agents that can write code, execute commands, browse repositories, and operate inside controlled runtimes. The workflow is remarkably close to the "Developer as Conductor" model: 🧠 Human Architect ↓ 🤖 Agent analyzes codebase 🤖 Agent writes code 🤖 Agent runs tests 🤖 Agent proposes changes ↓ ✅ Human approves 🎓 2. SWE-agent (Princeton University) 🔗 github.com/SWE-agent/SWE-age… Built by researchers from Princeton and Stanford. Instead of generating code snippets, SWE-agent behaves more like a software engineer: • Explores repositories • Reads files • Executes commands • Runs test suites • Creates patches • Resolves GitHub issues Its core innovation is the Agent-Computer Interface (ACI), which gives LLMs structured tools for navigating and modifying large codebases. 📜 3. AGENTS.md / CLAUDE.md This might be the most underrated practice in AI-native development. Modern AI coding tools increasingly rely on repository-level instruction files such as: • AGENTS.md • CLAUDE.md • AI_INSTRUCTIONS.md These files act as a constitution for AI agents. Examples: ❌ Never modify /core/architecture ❌ Don't add dependencies without approval ✅ Run tests before proposing changes ✅ Follow project coding standards Without explicit rules, AI learns from whatever code it sees. If the codebase is messy, the agent often amplifies the mess. The biggest shift happening in software engineering isn't better autocomplete. It's moving from: 👨‍💻 Human writes code to 🧠 Human defines intent 🤖 AI executes 👨‍⚖️ Human reviews The best developers of the next decade may spend less time typing code and more time orchestrating systems of agents. OpenHands: github.com/OpenHands/OpenHan… SWE-agent: github.com/SWE-agent/SWE-age… #AI #OpenHands #SWEAgent #SoftwareEngineering #ClaudeCode #OpenAI #AgenticAI #DeveloperTools #Programming

Congratulations to @NVIDIAAI on the release of Nemotron 3 Ultra, an exciting step forward in fully open language models. Nemotron Ultra was trained to work well with OpenHands, and we are happy to provide tier-1 support for it!

I couldn't help myself and had to dig into dynamic workflows. Check out my repo which explains it: github.com/rajshah4/workflow… Or a video if you want to hear me talk through it: youtu.be/PtbrKTgj3X8

Claude dynamic workflows seemed pretty cool and I wanted an open-source version, so I made one. This one improves the test coverage of your entire repo, but you should be able to swap out the prompt and do anything else: github.com/OpenHands/softwar…

We have also partnered with @MiniMax_AI to provide *free access to agents with MiniMax-M2.7* for a limited time! This is a great opportunity to test out how to mix in cost-effective models into your coding workflows. Try it out here: app.all-hands.dev

Come say hi to us at the @CAISconf in San Jose this week if you're around!

Hermes Agent now can orchestrate the @OpenHandsDev agents with a new optional skill! `hermes update` then do `hermes skills install official/autonomous-ai-agents/openhands` Reminder: You can already do this for claude code, codex, opencode, and hermes itself, you can force load the skill with `/<agent-name> <prompt>` or just ask hermes to use them, and it should find them. These ones are all built-in skills :)

In FOUR days, leaders from @OpenHandsDev @vercel @AMD @nvidia are going to be featured in a panel hosted by @alumniventures We've had over 100 sign ups in a week, spots are limited so head over to @LumaHQ to reserve your spot. RSVP link below

building a disciplined local coding agent setup... - Main surface: OpenHands - Local models via Ollama (7B-class on M3 Pro 18GB) - Codex will be the "strong external model as senior reviewer/verifier" - Will use strict modes mandatory structured handoff packets gates Anyone else running something similar with OpenHands local LLMs? What am I missing or what commonly goes wrong? #LocalLLM #OpenHands #AgenticCoding

Scaling Enterprise AI Agents event in NYC on May 26th! Speakers from @OpenHandsDev, @NVIDIAAI, @AIatAMD, and @vercel! If you're in the city then we'd love to have a good turnout.