Want to get into the top 10% of your engineering team by leveraging agentic engineering workflows? I'm planning a 2 day workshop here: https://form.typeform.com/to/EDuBEz4U

I shipped a feature with AI in record time.

Then I actually read the code.

60+ files. Weird abstractions. Tests that technically passed but proved almost nothing. Components rebuilt from scratch for no good reason. What looked like a finished feature was probably 70–80% of the way there.

And the worst part? This wasn’t my first time doing this.

I’ve also sat in a code review as a senior engineer and realized I couldn’t properly explain code that “I” had just written.

I use Claude and Codex constantly. I’m not going back to hand-writing everything. But after using AI to build software at startups, inside my own business, and on large production systems, I think we’re learning the wrong lesson from how fast these tools can code.

The question isn’t whether AI can write the code.

It’s how much you need to understand before you’re willing to ship it.