May 22, 2026
Too Prompt, Too Furious
Let the AI Cook
Coders are fighting over whether AI needs freedom or just better babysitting
TLDR: The article argues that AI writes better code when people stop over-controlling it and use simpler instructions. Commenters instantly split into two camps: one says that’s true, the other says the real problem is simpler—AI still messes up, and no amount of hype changes that.
A spicy little manifesto called "Let the AI Cook" has kicked off the kind of argument the internet was absolutely born for. The writer’s big claim is simple: stop smothering artificial intelligence with giant, fussy instructions and then acting shocked when it flops. In plain English, they say AI works best when it’s dropped into a familiar setup and guided by someone experienced enough to spot when it starts wandering off. Everything else? Fancy theater.
And wow, the comments did not nod politely and move on. One camp basically said, “Nope, this isn’t user error, the robot just isn’t that good.” That was the mood behind one of the sharpest replies, accusing the piece of being unfair to people who genuinely tried and still got bad results. Another crowd pounced on the article’s grand talk about “taste” and “judgment,” asking the obvious messy question: if those human skills matter so much, can anyone explain them without sounding like they’re inventing a new way to stay important?
Then came the internet seasoning. One commenter suggested playing mind games with the bot by asking leading questions instead of barking orders, like a therapist for code. Another dropped the gloriously petty line, “This is not even grammaring,” which is the exact sort of drive-by dunk that keeps comment sections alive. The whole fight boils down to a very 2026 panic: is AI making work easier, or is it exposing how much of people’s process was just ritual with extra steps?
Key Points
- •The article says success in AI coding depends mainly on the software stack and the developer’s prior experience rather than elaborate workflows.
- •It argues that long prompts with many constraints can reduce model performance by narrowing the solution space and embedding faulty assumptions.
- •The author presents a simpler two-step workflow: provide local reference context first, then state what to build.
- •The article says contextual priming helps the model infer conventions such as naming patterns, function structure, and error handling.
- •It concludes that developers’ higher-value work is choosing the stack, applying judgment, and detecting output drift, while much process overhead is described as busywork.