AI can ship a prototype; production is still the hard part
Anuradha Weeraman separates the first working demo from the much longer job of making software dependable, secure, maintainable, and supportable.
The essay argues that generative AI has compressed the path to a first working version, not the distance from that version to a production-grade product. That distinction covers integration, failure modes, operations, and ownership.
HN readers accepted the gap but challenged its use as reassurance: if review remains necessary, they asked, what new engineering workflow can keep up with the volume of generated code?
A demo has no pager
Anuradha Weeraman’s essay makes a narrow claim: AI can make a plausible first version quickly, while the work that makes a product trustworthy begins afterwards. A feature that appears to work still needs boundaries, integrations, monitoring, tests, recovery paths, and someone accountable when it fails.
The distinction matters because a prototype can hide those costs. A team that measures only the time to its first screen or endpoint can mistake generated code for a finished system.
The thread asks who can review the flood
The author put the premise plainly in the HN submission: > "AI has dramatically accelerated the path to a first working version. It has not shortened the distance between a first working version and something production-grade." — Anuradha Weeraman
One reader called this familiar advice a defensive response to the new pace, arguing that the real unresolved question is how engineers control a much larger flow of generated code. Another supplied a practical test: ask the model to assess whether the codebase is ready to be sold for $1 million, then inspect the gap between its claims and the result. The useful conclusion is not that AI cannot help; it is that teams need production checks that are as fast and concrete as their prototyping tools.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
Google News users say date filters no longer mean recent news
A columnist’s complaint about Google News became a thread about stale results, social posts in place of publishers, and the cost of losing a dependable news index.