Thoughts on engineering with AI in 2026

A few notes about using AI for engineering in 2026.

You can probably use AI in the most obvious ways and still get a lot of value.

A lot of my work in the past involved breaking tasks into small chunks meticulously and then letting coding agents run through them. This is still important! But, the dumb brute force ways can also be effective on specific types of problems. Things like “Find X random usages of old function, migrate them to Y, make CI pass, keep going until you’ve migrated at least 100” work. This was a nice surprise for me.

Verification is the bottleneck. You are the bottleneck.

Code generation is basically free now. The thing that makes you slower now is verifying whether the generated code is good. Things like CI are exceedingly important and need to be fast. Another thing that’s important is enabling agents so that they can verify their code more effectively, essentially keeping you out of the loop as much as possible.

It’s easy to fall into the trap of doing more instead of doing the most important thing better.

It’s tempting to have 5 different side projects running at the same time at your workplace in the hopes of creating more business value. It’s easy to spend less time on the strategy behind what you’re doing. AI helps do more work, it’s still your responsibility to make sure that the work you’re doing is actually worth doing.