There's something I couldn't code before that now takes 5 minutes. And I'm thinking about wrapping it in an MCP server to avoid the build pain.
The issue is project size. Incremental builds miss warnings. Full builds take forever. And when you have multiple agents working on the same codebase, they end up rebuilding constantly. The whole process becomes glacially slow.
A traditional IDE solves this. It watches your code during development, shows you warnings in real time, and maintains state about the build without you having to trigger it manually. An agent can't see that interface. It has to rely on console output, which is incomplete and slow to parse.
So I'm building a wrapper — an MCP server that can watch builds, capture warnings, and give agents real-time feedback without having to rebuild the whole project every time.
This is the kind of infrastructure that didn't matter before agentic development. When humans were the bottleneck, a slow build loop was an annoyance. When agents are the bottleneck, it becomes the constraint that prevents the entire system from working at scale.
The future isn't smarter models. It's better infrastructure for those models to work with. It's tools designed by people who understand what agents need.
And that work is just getting started.
Part of the #100DaysToOffload documenting agentic development in 2026