Can't you do this faster with AI?

I’m hearing this question asked a lot lately. Both within Mozilla and from others in the industry. You come up with a plan for implementing some feature, put your best estimate on how long it will take to implement, and then you get push back from folks several levels removed from the project along the lines of “Wouldn’t this be faster if you used AI?”, or “Can’t Claude Code do most of this?”.

This question bugs me for a couple of reasons. Firstly, because the answer is invariably “maybe”. I’ve been making use of AI tools for a while and while it does very well at some tasks, for others it doesn’t work as expected and the whole process would have been quicker to ignore AI entirely. I don’t think any of us are great at predicting when AI will work well; in many cases you just have to try, but also learn when to give up if it isn’t working out. I can’t currently base an estimate for a project on how much AI can do until I’ve spent a reasonable amount of time trying to get AI to do it.

Secondly, and more importantly, AI is just another tool. My strong opinion is that one of the most important things we do as engineers is evaluating the tools we have available and choosing the right tools for the project. In the past, while I probably had a rough idea of what tools I might use, I didn’t include that in the project plan (unless there is some specific need). Once the work starts we pick the tools and go. And this has always been fine. I’ve never once heard leadership or stakeholders respond to a project proposal with “Couldn’t you implement this quicker if you just used emacs?” or “Can’t you just use a binary tree for this?”.

But now AI is here and leaders are buying into the story that AI makes everything faster. Indeed, you can apparently build an entire web browser with AI in just a week or a C compiler for just $20k. These examples are impressive, but they’re carefully selected examples that generate buzz and generally don’t hold up to scrutiny or represent the messy reality of most software engineering. Yes, AI can be incredibly useful. But treating it as a magic productivity multiplier for all projects is misguided. We should trust engineers to evaluate whether AI is the right tool for their projects. If we could confidently estimate productivity gains from AI upfront, we’d already be factoring it in. Until then, let engineers do what they do best: pick the right tools for the job at hand.