AI Guides › Playbooks
By Nigel Guy · 5 min read
You have read enough headlines to feel that everyone else started in 2023 and you are now too far behind to bother. So you do one of two things: you keep reading about AI instead of using it, or you panic-buy five subscriptions and abandon four. Both feel like effort. Neither puts a real task through a real tool, which is the only thing that closes the gap.
The rule: Judge whether you are late by what people actually do with AI, not by how loudly it is discussed, then run one small, real task a week and write down what happened.
The loudest signal says "everyone is ahead". The measured one says something quieter. The Office for National Statistics published a UK business survey on 20 July 2026. It reports self-reported AI use among businesses with 10 or more employees rising from around 12% in late 2023 to around 35% by June 2026. Businesses with 0 to 9 employees were at 28%.
Now look at depth. Among businesses that use AI, only 10% describe their use as extensive. The average number of AI technologies per business moved only from about 1.4 to about 1.6. In plain terms, many people have tried something and few have built it into how they work. Note the caveats: these are self-reported answers from businesses, not individuals, and they say nothing about your particular job.
Anthropic's Economic Index, which analyses anonymised Claude conversations, points the same way: use differs a great deal by country, occupation and task, and people use these tools for explanation, documents and guidance rather than anything exotic. The opportunity is not a train that left. It is a skill most people have only sampled.
Three things, none of which is "AI moved too fast".
A ledger is a plain table you keep in a notes app or spreadsheet. One row per attempt. It turns vague worry into evidence.
| Column | What you write |
|---|---|
| Task | One job you already do, small enough to finish in 30 minutes |
| Tool | One tool only |
| Time without AI | Your honest guess, in minutes |
| Time with AI, including checking | What it actually took |
| Checked by me | What you verified or corrected |
| Verdict | Keep, tweak, or drop |
Choose something dull, repeatable and low-stakes: tidying meeting notes, drafting a polite reply, summarising a long document you own, turning a rough list into a structured plan. Avoid anything with legal, medical, financial or confidential content for now.
Claude, ChatGPT and Gemini all offer free tiers at time of writing, with limits that change; check each vendor's current page. Anthropic's help centre says the Claude Pro plan is $20 a month in the US (about £15 at time of writing; regional pricing varies and tax may be added, so check the £ price at checkout) and gives more usage than the free plan. You do not need a paid plan to run your first four rows. Upgrade only when the free limits actually block a task you have already proved useful.
The usual failure is a one-line request. Use a prompt like this:
You are a careful assistant helping me with [TASK_DESCRIPTION].
Context: I am [YOUR_ROLE] and the reader of the result is [AUDIENCE].
Material to work from: [PASTE_YOUR_OWN_NOTES_OR_TEXT]
Goal: [WHAT_A_GOOD_RESULT_LOOKS_LIKE, e.g. "a 150-word reply, friendly, one clear next step"].
Steps:
1. Tell me anything you need that I have not given you. Ask before guessing.
2. Produce the result in this format: [FORMAT].
3. List any claims or figures you were unsure about.
Constraints: British English. Use only the material I pasted; do not invent facts, names or numbers. Keep to [LENGTH].
Before you answer, check the draft against the goal and the constraints and fix any mismatch.
Fill in the square brackets with your task, role, audience, material, goal, format and length.
Read the output against your own material. Correct errors. Record the real time including checking. If the checking took longer than doing it yourself, that is a valid result: log "drop" and move on.
After four rows, read the table. Keep tasks where AI saved time after checking. Tweak ones that were close by improving the brief. Drop the rest without guilt. That is a working practice, which is more than most of the 35% have.
Hypothetical: Priya, a part-time bookkeeper, fears she is behind. Row 1: summarising a client's long email thread. Without AI, 20 minutes. With AI plus checking, 8, but she caught a wrong date. Verdict: keep, always verify dates. Row 2: drafting invoice-chaser emails. 15 minutes down to 6. Keep. Row 3: asking for tax advice. Dropped immediately; she has no way to check it and the stakes are high. Four weeks in, she has two kept workflows and a clear boundary.