A developer going by mohshomis built Modo, an open-source AI IDE that forces structure before code. Instead of jumping straight from prompt to code, Modo makes you work through requirements, design documents, and task lists first. The project is MIT-licensed and built on top of Void, a VS Code fork that had been sitting in maintenance mode.
It's opinionated on purpose.
The spec-driven workflow stores everything as markdown files in a .modo/specs/ directory. You get separate files for requirements (user stories, acceptance criteria), design (architecture, components), and tasks (implementation checklist). The agent works through tasks one by one, marking them done as it goes. If you close the IDE and come back later, you pick up where you left off, a form of session resumption that is crucial for long-term projects.
Modo also adds features Void didn't have. Steering files let you inject project rules into every AI interaction so you're not repeating yourself. Agent hooks can trigger actions based on events like file edits or tool use, useful for running linting or other checks. There's parallel chat sessions, subagents for independent tasks, and an autopilot/supervised toggle that controls how autonomous the agent behaves, an approach reminiscent of parallel agent swarms. The whole thing supports multiple LLM providers including Anthropic, OpenAI, Gemini, and local options via Ollama.
The approach is a deliberate contrast to tools like Cursor and Windsurf, which lean into fluid, conversational coding. Those tools work great when the AI gets it right, but they can leave you with spaghetti when they don't. Modo's spec files create a paper trail, making it easier to see where things went wrong and pick up later. Whether that structure helps or just adds friction probably depends on how much you trust AI to get things right on the first try.