The Boundary Moves

Chris Gregori, writing on his blog:

LLMs have effectively killed the cost of generating lines of code, but they haven't touched the cost of truly understanding a problem. We're seeing a flood of "apps built in a weekend," but most of these are just thin wrappers around basic CRUD operations and third-party APIs.

This is the right distinction, and Gregori draws it well. Code is cheap. Software — the ugly compound of maintenance, edge cases, data ownership, and everything that happens after the demo — is not. His comparison of AI-generated tools to spreadsheets is the best framing I've seen for what's actually happening: people are building disposable, personal software to solve immediate problems, then throwing it away. Not a SaaS revolution. A scratchpad revolution.

Where I part company is the implied permanence. The piece treats "AI can write code but can't architect systems" as a durable truth rather than a snapshot. It was true eighteen months ago that AI couldn't generate a working UI from a description. It was true twelve months ago that it couldn't reason across files in a codebase. The line between what's cheap and what's expensive keeps moving, and it only moves in one direction.

Gregori's real audience is engineers who need reassurance, and he delivers it competently. But the more interesting question — the one he names and then walks past — is what disposable personal software does to the mid-market SaaS businesses that survive on convenience and lock-in. The subscription tracker he uses as a toy example is, for a £15/month app with no moat, an existential problem wearing a side-project disguise.

The tools have changed, and the fundamentals of good engineering haven't — yet. That "yet" is doing a lot of work, and most commentary in this space pretends it isn't there.