Two months ago, I needed a new personal website and decided to give codegen a spin. Friends pointed me to Cursor and, hours later, I had a markdown website on a shiny new domain. Had I known hours earlier that Github hosted websites? Nope. How about a bio-inspired art thing for the entrance animation? Bang. Unnecessary but useful annotations on my CV page? Letsss gooo. Suddenly, I realized I could easily build my own software, and that this was true for everyone else.

In college, I first encountered the steep learning curve of syntax, methods, and functions. My classmates, it appeared, had been coding since they were 12 and flew through our assignments. Thoroughly deflated, I focused on hardware for the rest of my career, leaving software to others.

Cursor lodged dynamite in programming’s steep learning curve and blew it up in a giant, joyous explosion. Never in my life have I gone so quickly from thought to useful thing. The world of code is no longer off limits to me.

After building my website, I realized that I was a single wildebeest in a stampede. Daily, I’m meeting folks coding for the first time, dusting off out of reach app ideas. We all realized, oh shit, we can build software now.

I’ve been calling this software with a market of one. To solve a problem. To make software more personal. To have fun. What an awesome opportunity for us to use AI to become creators instead of consumers.

Some of this software will be terrible at first. I recently built a new app and it was bad. But building it gave me some new ideas that might actually be good. And, that little loop (idea -> it’s real! -> it sucks -> but it could be better!) can now happen very quickly, accelerating our ability to develop useful applications.

The first order impacts on tech seem clear. 25% of YC founders in the current batch said that 95% of their code base is generated by AI. Start-ups building software will have fewer people and need less capital. Products that serve TAMs too small for VCs, like software for life scientists, are much more likely to be built by small, bootstrapped teams. Start-ups are now distributing technology created by large companies by finding the niche industry applications. Software with programmatic interfaces, which I had always thought were quite strange, can suddenly be controlled by text or voice (example: Blender).

And, the role of enterprise SaaS companies is likely to change. If an employee needs to automate a workflow, they can build it themselves. And, the time saved from deleting the software procurement process is massive. Instead of SaaS, dev tool companies (code generation tools, models, etc) that serve employees or AI agents through a code gen interface will flourish.

Many work tasks can be completed using disposable micro-apps. Need a quick financial model for a business? Ask Claude Code to build one with sliders and dials for sensitivity analysis. Sometimes, it will simply be faster to build a tool than to hunt for the right pre-existing one. Given that the majority of coding time is waiting for codegen to finish, writing apps will get faster and faster.

Code generation will become a central aspect within products too. Want to customize the display of your oven? Ask it to do so and the oven will generate, test, and run the front-end code for your desired UI. Customization could be automated too with each person’s taste profile communicated to software products via API.

Mass customization probably won’t end with software. Code generation is the first breakthrough AI application, but it won’t be the last. Why won’t there be something similar for comics? Designing mechanical parts? Architecture? Fiction? Movies? It seems like it’s just a matter of time.