What makes an AI-built app slow

Compression first — it is a server setting, not a code change, and it typically cuts text assets by 70 percent.

Then caching headers on hashed build files, which stops repeat visitors re-downloading everything.

Then bundle size, which is real work.

Run a bundle analyser and look at the five largest dependencies before changing anything. The usual finds: a date library imported whole, an entire icon set, a charting library on a page with no chart, and a heavy dependency used for one function.

Then code-split so routes other than the landing page load on demand.

The download is the cheap part. Parsing and executing the JavaScript is what leaves a phone showing a blank screen.

Images without width and height cause the page to jump as each one arrives — which is why a visitor taps a button and hits something else. Set dimensions, or an aspect-ratio in CSS. On Next.js, next/image handles it.

Usually one of four things: a serverless cold start, a slow database query on the initial load, rendering on every request where the page could be static, or a deployment region far from your visitors. Work out which before optimising.