Quality isn't a single thing though.

Being reasonably sure you get the same thing when you order again is a kind of quality, which McDonalds has. So in that aspect, McDonalds provides a quality service. Not something I can say about my local kebab joint, where I have to consider who's working that day.

And the kind of quality that sets Maybach cars apart, is not something most people want to spend money on. Being able to produce a good-enough tool for a low price is one kind of quality, which many customers considers attractive.

That's where we're at at work. Our niche is serving a low-margin market, so customers are very price sensitive. They can tolerate some non-critical bugs here and there, if it means they don't have to pay as much.

If we were to follow Saarinen's path and only ship fully completed features or fix all bugs within 7 days, we'd most definitely would need twice as many technical people, which would significantly impact the price we'd have to charge.

That said, I think his other points, especially seeking to understand our customers and having small teams, is something that's worked very well for us as well. This has been aided by hiring folks from our customers in support roles, helping us to internalize domain experience.

While I agree overall with the author's thesis, they assume that prompting must be low resolution. It is possible to prompt with craft, it just takes more time; such is the cost of craft.

The irony is that AI makes craft more accessible to those who care deeply but lack traditional technical skills, while simultaneously making the absence of care more obvious than ever.

Software is not quality. That's probably why.