Software development with AI is starting to feel like cooking steak

Author left out the oven phase and importance of the input meat (which to a degree scale with price); probably explains his or her results.

Considering the article starts with "Cooking a steak requires almost no skill. Put it in a hot pan", I think I see the root of his problem. Pan searing a steak only makes sense in a small number of situations: you have a leaner cut like filet mignon, you want precise temperature control to achieve the Maillard reaction, or you know none of these things but just really like stinking up your kitchen and splattering butter everywhere for the aesthetic and the olfactory experience. For most cases, you would be better off with a grill rather than a pan.

Now maybe you're not knowledgeable about steak, and you're not sure whether the cut of steak you bought would be better off pan seared or grilled. Is ribeye better in a pan or a grill? That depends on whether you want it more smokey or more buttery. Ask an AI and tell it how you want it to taste, and it'll tell you the answer.

His metaphorical steak machine would produce better results if he had simply asked "I want a steak and I have this specific cut and I want it to have this kind of flavor, how do you suggest I make it?" rather than "cook steak on a hot pan, make no mistakes."

TL;DR: Tell the AI what you want and then ask how it for a recommendation of how to do it. You don't have to "understand software." You only need to understand how to ask the right questions.

It will be a push and pull and I'm curious where we'll be in a couple of years.

I can fly at 100mph if I let AI run loose and with a bit of steering I can get it to output what I'm looking for and generally pass verification and tests.

If I care about the code though, and I want to keep it maintainable, the amount of time and tokens I need to spend correcting and iterating on the output quickly eats through much of the initial time I saved, to the point where I'm unsure if I'm actually saving much time at the end of the process.

With hobby projects I lean on quality more, and the async nature of AI also makes this much easier to make progress without needing my full attention to do so.

In the corporate world, there's both the pressure to accelerate with AI, but also maintain code and product quality. The dials of one way or the other are more obvious now, but I don't believe it's possible to do both with the current models and harnesses without exponential cost.

What is interesting though, is I'm now leaning towards faster models rather than smarter ones.

Intelligence lets me bite off larger chunks of work at once, and trust the model to behave without having to watch it intensely, but doesn't seem to drive down the number of iterations required to hit my desired quality.

Faster models means the iterations I'll have to go through regardless will complete much faster and gets me closer to a proper flow state. Models will keep improving, but maybe we're getting near "smart enough" and the race will pivot to performance > intelligence.

To achieve the best outcomes, I agree.

But even a cheap steak can be decent if it's cooked properly. And an expensive prime cut can be ruined by cooking it improperly.

Agreed. I'm an avid cook. Sous vide circulators are cheap now, and so are meat thermometers. Doing a steak well is one of the easiest cooking tasks, to the extent that I won't bother paying top dollar in a restaurant because it's so easy to do restaurant quality at home. I could teach someone to cook a restaurant quality steak in a couple hours and a few practice runs.