Software development with AI is starting to feel like cooking steak
AI-assisted programming is likened to a “steak machine” that can reliably produce passable code, raising questions about whether businesses will settle for merely edible software instead of investing in expert craftsmanship. Commenters debate how much quality really matters when most users just need something that works, the economic incentives that drive a race to the “good enough” lowest common denominator, and whether human judgment, taste, and deep technical understanding will remain essential as models improve. Many see large gains for prototyping and routine tasks, but warn that overreliance on opaque generated code could entrench today’s buggy, bloated software rather than elevate it.
Overall reaction to the steak analogy
- Many like the framing that AI is a “steak machine”: it can crank out lots of passable output, but judgment and taste still matter.
- Others think steak is a poor analogy: cooking excellent steak at home is either trivially easy (with modern gear/technique) or far from the hardest dish; that undercuts the point.
- Some say the analogy nevertheless works because “simple to attempt, hard to nail consistently” describes both steak and software.
Quality vs “good enough” software
- Recurrent theme: the market often rewards “barely satisfactory” products; AI risks accelerating a race to the bottom in software quality.
- Several argue this mirrors broader economic trends (airlines, chips, consulting) and even government: lowest standard people will tolerate.
- Others push back: consumers also seek higher quality when they can afford it; multiple tiers (Ruth’s Chris vs Denny’s) will coexist.
How and where AI coding helps
- Strong support for AI as a huge accelerator for prototyping, glue code, CRUD apps, and “vibe coding” where correctness stakes are low.
- Many emphasize that most customers care about whether the product works, not how elegant the code is.
- AI is seen as finally delivering on promises of past “no‑code” tools (AppleScript, VBScript, IFTTT, drag‑and‑drop builders).
- Some report large personal gains in productivity and ability to build prototypes or entire side projects quickly.
Limits, risks, and the need for expertise
- Widespread agreement: to build good software with AI, you must still understand software—what to ask for, how to evaluate output, when it’s “charcoal.”
- Several liken LLMs to a “genius who gets drunk every 10 minutes”: better than most developers at local tasks, but lacking long‑range coherence and memory.
- Concerns about code bloat, lack of parsimony, and maintainability; AI tends to pile on special cases rather than design abstractions.
- Some fear a future of more buggy, opaque systems, with only a thin layer of experts able to diagnose failures.
Economic and organizational implications
- Expectation that businesses will cut developer headcount because “just edible steak” (mediocre software) is cheaper and acceptable to many.
- Others predict a boom in consulting/cleanup work as companies realize they’ve created unmaintainable AI‑generated messes.
- A few envision the future role shifting from “software engineer” to “user with taste” or “director” who steers agents rather than writes code.
Meta: article quality and AI discourse fatigue
- Several find the essay poorly written or “slop,” speculating it may itself be AI‑generated.
- Debate over the author’s use of “we” to generalize about developers.
- Multiple commenters express fatigue with constant AI pontification and long analogy‑driven thinkpieces.
Steak tangent (analogy stress‑test)
- Large sub‑thread on actual steak technique: sous‑vide, reverse sear, smokers, thermometers, deep frying, grills, etc.
- Disagreement over how hard it is to cook a “great” steak; some claim near‑foolproof methods, others note many restaurants still fail.
- This is repeatedly used to argue either that the analogy is apt (simple but subtle) or fundamentally flawed (too easy to master).