Blog

Entry-level job postings in the US are down 35% since early 2023, driven largely by AI handling the tasks that juniors used to do. Data entry, research, first-draft production, repetitive QA. The work wasn't glamorous. That was the point. It was where people learned to think like professionals.

The market is saturated with AI-generated work that looks almost right. Almost polished. Almost on brand. Good enough to ship, not good enough to remember.

The space race wasn't won by whoever announced it first. Several programs launched with bold declarations and more ambition than plan. What separated the ones that made it was years of unglamorous systems-building that nobody put on a press release.

AI platforms can now analyze heatmaps, scroll depth, session recordings, and click patterns before a single formal test is scheduled. Usability issues that used to hide until a researcher caught them in a lab are surfacing earlier and more consistently. For teams managing large products with limited research bandwidth, that's a real operational shift.

Most AI workflows start the same way. You stare at a blank prompt box, try to describe everything you need in one shot, hit submit, and spend the next hour correcting assumptions you never asked the tool to make. The coffee gets cold. The revision count does not.

In 1994, I was typing DOS commands into a black terminal screen to make a computer do what I wanted. An hour ago I was typing instructions into Claude Code's terminal to create a GitHub repository. Same black screen. Thirty years between them.