An autonomous coding agent living in the IDE - it plans before it codes, reads the codebase and follows call sites, runs commands and reacts to failures, and presents every change as a diff you approve. Extends through MCP. Live on the VS Code Marketplace.
The problem
Coding agents mostly live away from where engineering happens - in cloud dashboards and chat tabs - and the autonomous ones ask for trust before they have earned it. Wayne asks whether an agent can work inside the editor, with the developer holding approval over every file it writes and every command it runs.
How it works
- Plans before it codes: a request becomes an explicit approach, with clarifying questions asked up front, shown before any file is touched.
- Reads the relevant files, follows imports and call sites, and makes changes that match the conventions already in the codebase.
- Runs terminal commands and reacts to their output - installing packages, running tests, and fixing what breaks.
- Every edit is a reviewable diff and nothing is written without approval; operations you trust can be individually auto-approved.
- Extends through the Model Context Protocol to databases, internal APIs and other services, and works with Anthropic, OpenAI, Google and other providers on your own key.
Why it exists
It is the agents argument made concrete: autonomy is only useful as far as it stays inspectable. Wayne keeps the human in the approval loop at exactly the boundary where trust has not yet been earned - the same design rule I apply to production agent systems.
More work
Hiring for AI or ML?
I am open to AI/ML Engineering, Data Science, and Python roles, plus research collaborations and consulting. New York based, shipping worldwide.