Research. Product. Code.A personal website.

Good questions, loudly.

Ink illustration of Gareth Paul Jones speaking into a microphone, adapted from his photograph. — AI portrait after a photograph.

I work where research meets a real product. What should it do? How do we test it? Let’s build something and find out.

Start with the work

Selected work

Ideas, tested in public.

01 / Agents & evaluationVideo

How do you know
an agent is any good?

Test it against the work you actually expect it to do. A practical look at evaluating AI agents.

Watch: Evals for AI Agents

02 / Learning by makingEssay · 2016

And sometimes,
the answer is: stop.

“The feedback should have told me to stop, but as an engineer the drive to build something is often too great.”

From my face-filter app postmortem.

Read what went wrong

03 / A self-imposed deadline · 2015

30 apps.
30 days.

One mobile app, built and submitted every day. Source code on GitHub. A month of finding out by doing.

Read the experiment

A small patch of the internet.
Personal since 2002.

I’ve designed and built every version of this site. The stack changes. I still like making complicated things clearer — and taking them apart to understand them.

More about me

Good questions, loudly.

Selected work

How do you knowan agent is any good?

And sometimes,the answer is: stop.

30 apps.30 days.

A small patch of the internet.Personal since 2002.

How do you know
an agent is any good?

And sometimes,
the answer is: stop.

30 apps.
30 days.

A small patch of the internet.
Personal since 2002.