A lot of small teams have one or two people quietly using AI to write code — Claude, or tools built on top of it. It works. But the results live on someone's laptop, inside a chat session no one else can see. If it breaks later, or if someone else needs to do the same thing, you're starting from scratch.
OpenWork is an open-source tool that adds a shared layer on top of that kind of AI coding session. You run it yourself — no monthly subscription, no data leaving your office unless you want it to. When your developer (or the AI) does something, OpenWork keeps a live log: what was done, in what order, and why. It also lets you save common workflows as templates, so the next time you need the same thing done, you don't have to explain it from scratch.
Think of it like the difference between a chef scribbling a recipe on a napkin versus writing it up properly so anyone in the kitchen can follow it.
If you're paying someone to use AI tools to build or maintain things for you, this gives you visibility you didn't have before. You can see what actually happened, not just the end result. And if your team grows — or someone leaves — the knowledge doesn't disappear with them.
It's early, and a bit rough around the edges. But the idea is solid.
👉 Take a look: https://github.com/different-ai/openwork
Agentic session — when an AI isn't just answering questions but actually doing tasks: writing code, running checks, making decisions in sequence. It's working, not just talking.
Open-source — the tool is free and its inner workings are public. Anyone can inspect it, use it, or improve it. The opposite of a black box.
Workflow template — a saved set of steps that can be run again later, exactly as before. Like a recipe you've already tested.