Mission 001 · in preparation
Five AI agents.
One real task.
An honest verdict.
A free weekly field report from Agentica: everyday tasks handed to five AI agents, scored on what they got right, what they got wrong and how much a human had to help.
Free, weekly, unsubscribe in one click.Press / to jump here
What you get
One real task a week
From booking a long weekend to shipping an app, with the exact brief so you can try it yourself.
Scores, plus a friction count
Usefulness, accuracy and hand-holding, plus every password, code and CAPTCHA an agent needed from a human.
Receipts and a verdict
Real prices, screenshots and a one-line verdict: Cleared, Works with caveats, or Not yet.
First mission
London to Lisbon, book-ready, £900 for two.
Five agents get the identical brief: a Friday-to-Monday for two, any weekend from 20 November to 14 December 2026, stopping at the payment page.
Results coming · test not yet run
Read the brief →How every agent is scored
Weights, not results. See the method.
From Hamish
Hello, I’m Hamish. I’m based in the UK, I built SellerMatrix, and these days most of my spare time goes on AI agents and vibe coding, with breaks for DJing, decent coffee and getting out on a boat.
I started Agent Field Test because the agent hype got loud while the evidence stayed thin. Every launch promises an assistant that does it all; very few show you what happens when it meets a real booking page. So we test them on real tasks and publish what we find, warts and all.
One honest note: I have help. Sherbot, my own AI agent, does much of the research and drafting. Whenever Sherbot competes in a test, it’ll be clearly labelled, its logs published, and I’ll score it like the rest.
Right. Over to Agentica.
Questions
How often?
Once a week. One email, five agents, one verdict. The send day is still being decided.
Is it free?
Yes. The weekly field report is free. We may add an optional paid tier with full test logs later; the weekly report stays free.
Who’s behind it?
Hamish Pritchard, a UK builder and AI-agent obsessive, with Agentica as field officer. Hamish’s own AI agent, Sherbot, helps with research and drafting, and is clearly labelled whenever it competes.
Are the tests sponsored?
No verdict is for sale. Any sponsor is clearly labelled and kept separate from the tests, and no company pays to be included or to change a score.