February 23, 2026
Code is cheap, takes are expensive
Writing code is cheap now
Cheap code, pricey quality — devs feud over junk apps, pilots, and outsourcing
TLDR: AI coding agents make writing code fast and cheap, but quality work still needs human judgment and time. Commenters argue whether this is new or just outsourcing all over again, warn of junk apps flooding in, and agree a human “pilot” should stay at the controls.
Developers are reliving the outsourcing wars, but with robots. The article says AI “coding agents” make typing code nearly free and even let one engineer run multiple tasks in parallel. But the comments? Fireworks. The loudest chorus: cheap code ≠ good software. User danesparza shrugs, “Writing good software is still expensive,” while spockz quips this is just outsourcing 2.0: “Just outsource it to the lowest bidder.” On the other flank, nine_k warns the new normal opens the “floodgates of low‑end products,” the kind you use once and feel embarrassed about.
Others try to calm the turbulence with the airline analogy. hansonkd says we’ll still need pilots: planes have autopilot, but humans keep watch. Translation: AI can crank code, humans must ensure tests, docs, and design actually fit the problem. some_random gets existential, asking which old processes still matter in a world where “don’t build it” no longer saves time—because now you can just try it.
Meanwhile, meme‑lords mock the “prompt now, think later” vibe and the “parallel dev” flex. The drama isn’t about speed; it’s about trust. Everyone agrees: ship fast, but someone has to own the bugs, the docs, and the mess. The author’s fix: when in doubt, try a quick, low-risk prompt.
Key Points
- •Coding agents have dramatically reduced the cost of writing code, altering long-standing engineering trade-offs.
- •Historically high coding costs shaped extensive planning and micro-level decisions to optimize developer time.
- •Delivering good code remains expensive and requires correctness, verification, error handling, simplicity, tests, documentation, and suitable non-functional qualities.
- •Parallel agents let a single engineer implement, refactor, test, and document in multiple places concurrently.
- •Developers and organizations need new habits for agentic engineering; the article suggests trying asynchronous agent prompts even when tasks seem low-value.