Software development with AI is starting to feel like cooking steak, engineer writes

A software engineer argues that AI has made building software easier without making it reliably good, in a blog post published in July and titled "Almost No Skill Required to Cook a Steak (Though You Probably Can't Make a Decent One)". The author, Yurii Sydorets, writes that cooking a steak requires almost no skill — put it in a hot pan, wait, flip it, and you get something technically edible — but a genuinely good steak, medium-rare from edge to edge and consistently delicious, is a different matter. Software development with AI, he says, is starting to feel much the same.

Building nonstop

The post describes how developers now "build nonstop": with AI, without AI, during the commute — creating agents, tools, prompts and feedback loops, then throwing everything at a model in the hope it returns what they imagined, "without ever having to understand how any of it actually works." What people want is the perfect steak: software that works, looks good, feels polished and arrives exactly as imagined, with the same result every time. What they get, Sydorets writes, is not even close — sometimes something surprisingly good, other times "charcoal with a sprig of thyme on top" presented confidently as medium-rare.

Why "going to a restaurant" doesn't fix it

The usual response is to pay for the problem to disappear: a premium AI product, an agency, another coding assistant, a new framework promising professional results. Sometimes someone else has solved the problem, the author notes; quite often they have not. His argument is that AI is not a chef: "At best, it's a steak machine." It can follow a recipe — watch the temperature, flip at the right moment — and repeat it fast at enormous scale. What it cannot do is know what you want unless you translate it into requirements, constraints, examples, tests and feedback, and even then it is limited by its capabilities, its context window and the quality of the system built around it.

In his analogy, even the expensive restaurant ends up serving the same burnt steak, because "every restaurant in the city hired the same AI cook" in the name of cost optimization — and most customers will not notice, because most software only has to be acceptable. The people who notice are those who actually wanted to make the thing.

The conclusion: learn to cook

Sydorets's answer is not to abandon AI but to rely less on luck. AI can make developers faster by automating repetitive work, producing a starting point and explaining code, but it cannot replace judgment: it cannot define quality for you, decide which trade-offs are acceptable, or always catch the moment when something is technically correct but wrong in every way that matters. "To build good software with AI, you still have to understand software," he writes — to know what you are asking for, how to judge what comes back, and when the machine is "confidently serving you charcoal."