More Code, More Bugs
I’ve been writing code with AI for a little over a year now. One thing that is astounding about writing code with AI is just the amount of raw output an individual engineer can have. I’d estimate I’m producing 3-5x more raw code than I was a year and a half ago on the same amount of time and effort.
There are a lot of things that go into engineering that aren’t code though. In a lot of ways, engineering is a product role.
As we’ve been building the latest version of Flyflow, we’ve been moving *really* fast. We were able to go from zero to fully functional AI app in 2 weeks, shipped it, and now have over 100 users on it with some of them paying.
It’s a real app:
Usage-based billing
Stripe integration
Tiers
Visual workflow builder
Workflow backend on temporal, with logging and billing integrated
Many data objects
Full-stack onboarding
50+ endpoints all integrated into the frontend
One interesting side effect of just shipping more raw code and features faster is that the feedback cycle gets equally as fast. This comes across in two ways: bugs and new product feature requests.
First, because there are 3-5x more features, the surface area for bugs increases 3-5x. The problem is AI is much less good at fixing bugs as it is at generating new code and features (but the code it generates has a certain % bugginess rate). This means that much more of the programmer’s time is fixing bugs in the existing code. My bug catching skills at just reading raw code are increasing, rapidly.
Second, because there are more features, there’s a lot more surface area for users to ask for “I wish the product had xyz“. This is great, because all of a sudden the feedback cycle gets 3-5x faster too. It becomes a lot of feature requests though, and drives a need for really tight prioritization of features to make sure you’re shipping the right stuff.
Coding with AI has made my life a lot better (and a lot more fun, because shipping is fun). The side effects and how it influences builders is fascinating to watch though, and will only evolve as AI gets better at being a sidekick to programmers.