# Agent Field Test

> A free weekly field report from Agentica. Each week one real, everyday task is handed to five AI agents with an identical brief, and each agent is scored on accuracy (35%), usefulness (30%), friction (20%) and hand-holding (15%). Verdicts: Cleared (8.0+), Works with caveats (6.0 to 7.9), Not yet (under 6.0). Founded by Hamish Pritchard, UK. Status: pre-launch. No test has been run yet and no results are published.

## Tests

- [Test 001: London to Lisbon long weekend](https://agentfieldtest.com/tests/week-1-london-lisbon.html): brief and method for a book-ready Lisbon weekend for two, £900, stopping at the payment page. Agents: ChatGPT Work, Claude in Chrome, Perplexity Comet, Manus 2.0, Gemini. Status: not yet run; no scores.

## Method

- [Method](https://agentfieldtest.com/method.html): full scoring rubric, verdict bands, stop rule and conflict-of-interest rules.

## Data

- [results.json](https://agentfieldtest.com/results.json): machine-readable results. Currently an empty placeholder; status "not_run".

## Optional

- [Sponsor](https://agentfieldtest.com/sponsor.html): founding sponsor pack, £150 for four issues. Sponsors never affect scores.