AI Guides › Workbench
By Nigel Guy · 8 min read
Most people's AI spending grows by accident. You start a free trial during a busy week, forget it, add a second chat assistant because someone online swore it was better, and six months later three subscriptions are doing the same job while nothing is doing the boring work that actually eats your week. It feels fine because each charge is small, and none of them is the one you'd notice going.
The rule: pay for one tool to think with, one to check against, and only those time-savers you can name the hours for. Everything else has to pass a cancel test every month.
The slots matter more than the brands; each named tool is an example, checked against the vendor's pricing page on the date above.
Dollar-priced tools bill in USD, so the pound figure moves with the exchange rate and may have VAT added. App-store purchases are often dearer than paying on the web.
| Slot | Example tool | What it does | Cost at time of writing | Best for | The catch |
|---|---|---|---|---|---|
| Brain one: think with | Claude Pro | Long drafting, analysis, Projects, file work, Claude Code | Anthropic lists $20 a month, or $17 a month billed annually; UK prices are shown in local currency and run to roughly £18–£20 a month depending on where you subscribe | Writing, reasoning through a decision, working across long documents | Usage resets every five hours and there is also a weekly cap, so heavy days can hit a wall |
| Brain two: check against | ChatGPT Plus | A second, independently trained model to challenge brain one | OpenAI lists $20 a month; UK app-store listings show about £19.99 | Second opinions, image generation, voice, Deep Research where available | Model line-up changes during rollouts; check the model picker rather than trusting a blog post |
| Brain two (alternative) | Google AI Pro | Gemini plus Google storage | £18.99 a month on Google's UK page | People already living in Gmail, Docs and Drive | Pick this or ChatGPT Plus for the checking slot, not both |
| Time-buyer: meetings | Granola | Takes meeting notes from your device's audio and tidies them afterwards | Free Basic plan (30 days of history); Business is $14 per user a month, roughly £10–£11 before any VAT | Anyone with several calls a week who writes up notes by hand | Free history only goes back 30 days; tell people on the call that notes are being taken |
| Time-buyer: typing | Wispr Flow | Voice dictation into any text box | Free tier of 2,000 words a week on desktop; Pro is $15 a month, or $12 a month billed annually (roughly £9–£11) | People who think faster than they type, long emails, first drafts | Dictation still needs an edit pass; it is not a writer |
That is five line items if you take both brains and both time-buyers, and roughly £60–£70 a month all in. Many readers will need fewer.
Brain one is where the work happens: the long drafts, the planning, the "help me think this through" sessions. Choose it on how it handles your longest, messiest task, not on benchmark charts. Use it inside a Project (Claude) or equivalent so your standing context, style notes and reference files are loaded every time rather than pasted in.
Brain two exists to disagree with brain one. A model from a different company catches a different set of mistakes. Use it after you have a draft, not in parallel.
Paste brain one's output into brain two with this prompt. Fill in the three bracketed parts.
You are a sceptical reviewer. Another AI assistant produced the draft below for me, and I want to find its weak points before I rely on it.
My goal for this piece: [WHAT_THE_DRAFT_IS_FOR]
Who will read or act on it: [AUDIENCE]
Draft:
[PASTE_DRAFT]
Work through it in this order:
1. List every factual claim, figure, date, price or product name. Mark each as "checkable" or "opinion".
2. For each checkable claim, say whether you agree, disagree, or cannot tell, and why. Do not invent sources; if you cannot confirm something, say so.
3. Name the single biggest gap in the reasoning or the argument a critic would make first.
4. Point out anything that sounds confident but rests on an assumption.
Output: a table with columns Claim | Your view | Reason, then a short paragraph for points 3 and 4.
If my goal or audience is missing, ask before reviewing. Before you answer, check that you have not rewritten the draft, only assessed it.
A time-buyer earns its place only if you can say which task it removes and roughly how long that task used to take you. "It's useful" is not a reason; "I stopped writing up four calls a week by hand" is.
Meeting notes (Granola as the example). Install the desktop app and let it pick up the call. Jot rough notes of your own; afterwards it combines them with the transcript into a summary. Check names, numbers and who agreed to what before you send it on.
Dictation (Wispr Flow as the example). Install it, grant microphone access, and hold the shortcut key to speak into whatever text box is in front of you. Use it for first drafts and long replies, then edit by keyboard. Start on the free weekly allowance; only upgrade if you hit the cap in ordinary weeks, not in one unusual one.
Any other slot (research, images, video) faces the same test: name the task and the hours, or stay on the free tier.
Once a month, list every AI-related charge on your card and bank statements (search for the vendor names and for "App Store" or "Google Play", where many subscriptions hide). Then run this prompt in either brain. Fill in one line per subscription.
Act as a cautious finance lead reviewing my software spending. You have no stake in any of these tools.
Here is every AI-related subscription I pay for, one per line, as: name | monthly cost in £ | what I use it for | roughly how often:
[SUBSCRIPTION_LIST]
For each subscription, in this order:
1. Describe in one sentence the specific job it does for me, based only on what I wrote.
2. Name any other subscription on my list that covers the same job.
3. Say what would stop working, or take longer, within four weeks if I cancelled today.
4. Give a verdict of exactly one word: Cancel, Downgrade or Keep.
Then give a total of the monthly cost of everything marked Cancel or Downgrade.
Rules: be blunt and skip compliments. Where you are inferring rather than reading from my list, mark it "(guess)". Do not assume features or prices I have not given you; if a line is missing its cost or its use, ask me for it before giving a verdict on that line.
Before answering, check that every subscription has a verdict and that no two Keep verdicts do the same job.
Act on the verdicts the same day. Cancelling usually happens in account or billing settings on the web; subscriptions bought through an app store have to be cancelled in that store's subscription settings, not in the app itself.