Check an AI announcement

An announcement is a reason to investigate a capability, not evidence that a tool will handle your work. Compare a specific claim with primary documentation and a task you can test.

Make a claim record

Claim from the announcement:
Primary source URL:
What the documentation says:
Conditions or limits:
Task I want to test:
Input and expected output:
What would count as failure:

Check what the result actually shows

A benchmark, staged demo, and result on your own task answer different questions. Record which you have. Keep an untested claim separate from an observation.

Read the primary source

Examples include the Agent Skills specification, the Hermes repository, and Google’s TurboQuant research post. Record what each source actually supports before applying a headline to your own work.

Use a small comparison

Try the same input and acceptance checks with your existing process and the proposed tool. Include missing information and a conflicting input. Record review and correction effort, not just the first answer.

Start with the action-list exercise if you need a task with an answer key.