The only way a company grows that fast (by pure word-of-mouth) is if the market is still extremely young with tons of green field opportunities. Recently, yours truly converted and paid his first $20, contributing to their incredible revenue growth.
I haven't worked on any code seriously since building Fin back in 2020. When I recently started a new project to automate my wife's monthly reporting (she's a property manager), I decided to download Cursor to experience the revolution for myself. And yes -- it is pure magic. Within hours, I had a completely working script that automatically ingests data in various formats; cleans, analyzes, and formats that data; and then crafts a beautiful Excel sheet in exactly the design my wife wanted for her monthly statements, ready to distribute to her clients. Before AI, simply reading the pyxl library documentation to read and generate Excel files would have taken me hours.
The last time I was blown away by this kind of PURE MAGIC was when I compiled and ran my very first "Hello World" in C++ in 2003.
However, while magical, it also opened my eyes to all the things that aren't perfect, haven't been built, and that will need to be iterated upon over the coming years. Specifically:
- AI is terrible at understanding and maintaining code architecture. If I didn't manage Claude like a hawk, its generated code would become spaghetti within just a few turns. It loves creating duplicated functionality, badly named variables, incoherent function definitions, and convoluted (and error prone) logic.
- AI makes a lot of assumptions. Today, it isn't in the habit of clarifying and thinking through a user's requests, it just EXECUTES. That means that many times it misunderstands or makes assumptions about how I want things to work in the underlying logic, which can create extremely difficult to find & fix bugs if I'm not careful.
- AI is overly confident. It's common for it to believe it knows how to do something -- and implement it -- only for it to be totally and completely wrong.
- AI doesn't understand the context. When implementing new functionality, it doesn't remember what hard coded assumptions it made just a few minutes ago, leading to broken code that quickly devolves into mush.
All this to say, I think we're literally on version 0.1 of what coding could be, and while Cursor is truly an amazing product, it's also obviously incomplete. There are tons of opportunities for Cursor and its competitors to evolve its product offerings, and if Cursor isn't careful, another company may end up stealing its market overnight. After all, Cursor still hasn't built any products or features that disincentivizes me to switch to the next shiny tool. It'll have to do so soon to keep its crown.
Startups AI
This post was originally shared by on Linkedin.