Meet AgentDesk: a learning environment for AI-assisted development

AgentDesk is an open-source tool I've been building as part of my own AI learning journey. The idea is simple: take the best ideas I've found in the AI space and put them into one tool, so new developers have one place to learn how AI coding works and how its development workflow differs from traditional coding. It runs locally as a small AI dev team. A planner breaks work into tasks. Developer agents take those tasks, each with its own role and LLM model. A reviewer checks the work, and you decide what gets merged. You can watch the whole cycle happen in an isometric office view, so it's easy to see who is doing what.

Under the hood it pulls together a few ideas I've learned a lot from:

• Vercel-style UI methods for a clean, modern interface
• Google's development skills and practices
• Karpathy's take on LLM memory, built as a project wiki that the agents read and keep up to date
• Worker nodes with different roles and different LLM models

Who I hope it helps:

• People with less coding experience, so they can catch up to more complex coding tasks quickly without losing hope along the way
• More senior developers, so they can see how AI fits into a modern development cycle and make that transition with confidence

What's coming next:

• Docker deployment
• Team features like planning, plus application architecture documents with generated diagrams
• Data factory, which lets companies and developers test different LLMs on their own data, systematically. It will compare and report on LLM performance, accuracy, cost, and quality against a human baseline.

It's available here: Agentdesk in Github. I'd love feedback, ideas, and people to try it, especially if you're early in your coding journey or helping a team move to AI-assisted development.