AI agents helped me build faster, but made the workflow harder to manage

AI coding agents made me faster, but they also made my workflow messier.

I’ve been building a real app with them and the biggest lessons were not only about prompting. I had to learn a lot around API integration, Cloudflare, storage, scaling costs, and why moving from a web app to Electron can make sense when the product needs local files, terminals, and a desktop workflow.

Cloudflare was a good example. At first I saw it as “DNS and protection”, but while building I started to understand the bigger reason people use it: predictable storage, routing, caching, and long term scaling without turning every read/write into a future cost problem.

The other thing I noticed is that agents still work like isolated workers. One edits something, another reads files, another waits for approval, and suddenly you are not just building the project anymore. You are managing the workers.

So now I’m more interested in the workflow around AI agents. How do we see what they changed, what they opened, where they got stuck, and how they should share context?

For people building with AI agents, what has been the bigger challenge for you: getting good code from the model, or keeping the whole project coordinated?