I recently wrote an article for LeadDev about how software engineers are dealing with the AI hype in practice. Here’s my position, in short:

I think AI is pretty great. It speeds me up¹ by about 50% overall - in practice, sometimes 0% and sometimes 300%, depending on what I’m working on. I have to do gnarly legacy changes myself, since there’s way too much wicked context for the AI to consume. But if I’m making a change in a small, fairly isolated codebase, Copilot Agent mode can get me 90% of the way there.

The problem is that tech leaders are expecting a consistent 10x speedup, not the 1.5x speedup I’m getting. When they vibe code things, they’re always working on greenfield projects where AI is most effective - or they’re not doing the last 10% that you’d actually need to get a brownfield change into production. It’s hard to explain that AI still can’t do the hardest parts of the job.

And I’m having a comparatively easy time, since AI works well for me. My colleagues who are AI skeptics or dislike it on moral grounds are under much more pressure. They’re fighting on two fronts: against leadership’s new AI mandates, while also struggling to review the tide of partially AI-generated code that their AI-friendly colleagues are generating. There’s a lot of internal conflict in the industry right now, which is especially tough now that layoffs are much more common.

If you want to read much more about this, you can check out the article. But I wanted to note my position down in a more unpolished way here.