> Sure, but the harder part is measuring whether the developer is actually 5% faster. Otherwise you can make the same case for every $10/mo subscription service in the world, and so we should all be operating at infinite efficiency.

After n such iterations, the developer gets 100*(1-0.95^n)% faster. So, after some such $10/month purchases, the developer gets so fast that buying another improvement yields diminishing returns.

I agree. I can see these AI assistants becoming a game changer and eventually a requirement to keeping up, but the costs will be prohibitive for engineers in many regions.