Choosing your daily driver — a fair test

Nine lessons of tools, and one question left: which one is yours? Not “which is best” — you’ve seen why that question has no stable answer — but which deserves the browser tab you open by default. Rankings can’t answer this, because rankings don’t do your job. A twenty-minute test can.

Build your five-task benchmark

Pull five tasks from your actual recent life — the L1 note you made about what you already use and what annoys you is the perfect quarry. A good set covers different zones, for example: rewrite a real email for tone · summarise a real document and cite where key points came from · a factual question from your field, sources required · brainstorm openings for a real problem · explain something you half-understand, then answer your follow-up.

Write each task down once, exactly, with the material attached. Same words for every assistant — fairness lives in the wording.

Score on a rubric, not a feeling

For each output, three quick scores (1–3): Useful — could you act on this with light edits? Faithful — accurate to your material and verifiable on specifics (the E1·L7 checks apply during the test — verifying is part of scoring)? Effort — how much fixing or follow-up did it take? Fifteen numbers per assistant. Twenty minutes for the whole exercise with two or three contenders — the same blind-scoring idea our Comparison Lab will automate, done by hand today.

Read the results like a pro

Expect a split decision: one tool wins your documents, another your research, another lives where your work does (L4/L6’s integration gravity). That’s the normal, correct outcome — a default plus specialists beats a monogamous choice. Write one sentence: “X is my default; Y for long documents; Z when I need sources.” Congratulations — you have an AI toolkit, chosen on evidence.

Then diarise a re-run in six months. Strengths move (L2–L8 said so on every card); your benchmark is how you notice without living on tech news.

Check your understanding

1. Why must every assistant get identical task wording in your benchmark? Assistants share answers otherwise Different wording tests your prompting variance, not the tools — fairness lives in the wording Vendors require it

2. Your test splits: one tool wins documents, another wins research. The professional conclusion is: The test failed — rerun until one wins everything Adopt a default plus specialists, and note which task goes where Pick whichever has the nicer logo

Recap

Five real tasks, identical wording, a 3×5 rubric, a default-plus-specialists verdict, re-tested twice a year. Module E2 complete — you now choose tools the way this platform reviews them: on dated evidence, not brand gravity. Next module: prompting fundamentals — multiplying what every one of these tools gives you.