(Illustration by Public domain vectors on Unsplash)

When it comes to vibe coding and other AI code-generation efforts, I think about icebergs.

The "iceberg" analogy – in which only a portion of a concept is visible, with a larger mass underneath – is somewhat cliche. But this analogy comes up a lot because it's so widely applicable.

People who don't work in the software business might reasonably assume that the field is all about code. Code is the only part of software development that they hear about. For them, that's the visible part of the iceberg that sits above the water.

Below the surface, a proper application development effort also involves roles like ops and QA and product management. It's further supported by a host of best practices that keep the entire effort on the rails. These subsurface elements form a system of checks and balances around the code to ensure that the company delivers a quality software product to the market.

This gets to why AI code generation works so well in some situations and stumbles in others. In the hands of a sufficiently experienced software developer who knows how to delegate tasks and who embodies industry best practices, code generation tools truly are a force multiplier. A company that supports such a developer with product, QA, operations teams can dramatically speed time-to-market in a way that is still reasonably risk-managed and future-proof.

Outside of that scenario – any time you have a person writing code without that support – you're rolling the dice. Maybe you get an app that works. Maybe you get an app that runs but only appears to actually work. Time will tell.

The takeaway lesson: AI code-generation tools certainly improve some aspects of software development. But those tools alone won't replace an entire dev team.

(This may all sound familiar to some of you. I've been saying this a lot as of late, so I figured I'd share it here for easy reference.)