Originally posted on LinkedIn on July 28, 2026
Over the past six months, I’ve been creating my own informal AI benchmarks.
I’ve followed AI progress closely for about four years. From professional benchmarks in language, coding, maths and science, I already knew that the models were becoming remarkably capable.
But benchmark scores are abstract. Watching AI perform well in your own professional and personal domains hits differently. You don’t just understand the progress intellectually, you experience it.
Some things I’ve created with ChatGPT and Claude this year, with my own working time:
- Animations for five children’s stories (~24 hours)
- A simple Snake game (~4 hours)
- A personal weather app with rain radar (~12 hours)
- Four short comedy videos (~16 hours)
- A sculpture portfolio website (~8 hours)
- A personal recipe library (~4 hours)
- My quarterly tax filing (~1 hour)
- A professional interface for an enterprise web app (~6 hours)
The last example was particularly interesting.
UI/UX design has been my profession for nearly twenty years. For a recent project, I gave Claude Fable 5 the functional design, design system and existing mock-ups, discussed the approach with it, and had it produce the first interface designs directly in HTML.
It worked much faster than I normally would, and the result was at least comparable to what I would usually deliver as a first design pass.
My expertise still mattered: understanding the problem, providing context, making decisions, evaluating the output and steering the model. But the amount of execution I could delegate was really impressive.
My job hasn’t disappeared, but it has changed permanently.
Has AI maxed out all my personal benchmarks? Not yet. It still struggles with Dutch cryptograms, and an attempt with my father to build a bridge-playing program failed.
But I wonder for how long.
What are your personal AI benchmarks—the tasks you regularly retry to see how much the models have improved?
Back to AI experiments and musings