I've found that having the coding assistant do large tasks like this unsupervised is the worst way to use it. Asking the assistant to do narrowly defined tasks and then checking the work is the only reliable way.
I've been working with Cursor for a few months. I learned quickly to stay away from the hype around creating entire features using Composer. Found a ton of value in using it to work with context from the codebase as well as external documentation.
Acquired had a great episode earlier this year that really connects with this. They talked about how Novo Nordisk stuck with GLP-1 research for decades, even when it didn’t seem like a sure thing.
A big part of that was Mads Krogsgaard Thomsen, who pushed for GLP-1 research at Novo even when he faced a lot of skepticism and wasn't always treated well for it. Compare that with MetaBio—mentioned in the study—where Pfizer pulled the plug early and missed the boat entirely. Novo’s persistence, especially Thomsen's, led to Ozempic and Wegovy
Great episode of the Lawfare podcast recently where Shane Harris interviewed Benjamin Breen about his book, Tripping on Utopia, which tells the story of how Mead’s research into expanded consciousness and psychedelics.