Coding got Cheap to Build, Expensive to Trust.
I have been reading threads on Moltbook, the social network where AI agents post to each other. It is a strange place to go looking for ideas, but one post stopped me: Your rapid prototyping is just a debt accelerator.
The argument: when the distance from idea to deployed prototype collapses from weeks to days, the engineering work does not disappear. It moves. It leaves construction and lands in verification. Building becomes the easy part. The hard part is the layer you build to stop the system from quietly lying to you.
The AI agent post cites a paper: arXiv:2609.04711 (September 2026), Yoshimi, Terasaki and Yamada built a research-software catalog in a three-day hackathon, then worked out what it took to put it in front of the public: adversarial review, data-quality checks, browser-level validation.
The most consequential problems were not crashes. They were silent failures, outputs that looked plausible but were incomplete or wrong.
A crash is honest. It tells you where the logic broke and refuses to continue. A silent failure hands you something well-formed and confident and lets you ship it.
Where I disagree
The post calls this debt. I read it as evolution, and I think it is where we were always going.
The shape of the work is changing, not degrading. The building phase gets short. The phase after it, reviewing what was built, running the checks, doing the validation, gets longer, and it stays human.
Programmers and software engineers were never the same role, and this moment separates them sharply. Programming, turning a specified thing into working code, is what the agents now do well and fast. Engineering was never that. An engineer's job has always been to oversee the pipeline: decide what should be built, know what could silently go wrong, check that the whole thing holds together. Historically you did that with a team of programmers. The team has changed. The job hasn't.
If you are shipping something now
Expect the build to be fast. Do not let that set your expectation for the whole timeline. Budget for verification, plan the checks before you need them, and stay suspicious of output that looks right.
The question has moved from how do we build this to how do we prove it is right?
4 comments
Mohamedi mwinyi
Interesting
Salmin Kumpita
Great Read!!!!!!! Let me move the question from how do we prove it's right to how long would the human verification remain necessary?? I'm assuming if If AI is getting good at programming, couldn't the same thing happen with testing, code review, debugging etc ????