OpenCode 2.0
OpenCode 2.0 lands in the terminal with a simple pitch: the coding agent is open source and yours to run. That matters because most AI coding tools are walled gardens where you trade your workflow for a subscription and whatever model the vendor chooses. With OpenCode, the agent itself is the product you can inspect, and you can swap in different models depending on the task or your budget.
I think the terminal is the right home for this. Developers already live there, and an agent that can see your git history, run your tests, and edit files feels more like a teammate than a chat window. The open source angle is what gets me excited though. It means the community can audit what the agent does, fix its blind spots, and extend it without waiting for a roadmap. That's how tools like this get trust.
What I'd want to know next is how well it handles real-world repos with messy dependencies and long build times. The demo is easy; the daily grind is the test. If OpenCode can survive that, it could become the default way a lot of teams work.
