What I'd tell a founderspending £600 a year on AI tools.

Roughly twice a week, a founder messages me asking if [tool X] is "worth the £40 a month." Almost every time, the honest answer is no — not because the tool is bad, but because they are paying for it three quarters before they need to.

What follows is an unsentimental audit. It is not a list of "the best free alternatives" — that genre is now mostly affiliate slop. It is a framework for deciding, tool by tool, whether the spend is doing work, and where the same money would buy more output.

The numbers below are real. They are from the spending audits I've run with founders over the last six months. The average annual AI spend on this cohort is £620. The average amount that was actually moving the business: £180.

The audit (run this monthly)

Open your card statement. Pull the last three months. List every AI-flavoured subscription. For each, answer four questions in one line:

1. What specific job did this tool do this month that the free tier of [your default AI] could not?
2. If I cancelled it tomorrow, what would actually break?
3. Am I using more than 25% of the feature surface I'm paying for?
4. Would I notice if it became 30% slower or 20% less accurate?

If the answers are "nothing specific", "nothing", "no", and "no" — cancel. The tool is on the books because you signed up during a launch, not because it earns its keep.

The four most-cancelled categories (in my audits)

1. The "AI writer" tools. (£15–£40/month.) Almost universally cuttable. Whatever the tool does, ChatGPT/Claude does at least 80% of, on the free tier. The "AI writer" was useful in 2023 when the underlying models were locked behind APIs. They are not now. If you bought one for "SEO content generation" — see post 3 in our blog on why prompts beat tools.

2. The "AI assistant for [niche]" wrapper apps. (£20–£50/month.) The wrapper does prompt engineering and a UI on top of GPT-4 or Claude. If the prompts they're running are good, you can copy them. If you can't see the prompts, you're paying £30/mo for someone else's prompt library. A one-time prompt bible at £47 beats a SaaS subscription at £360/year by a wide margin.

3. Auto-content generators for social. (£25–£100/month.) Either generates obviously-AI threads (which cost engagement long term) or requires so much manual editing that you saved no time. Cancel. The output also drives the very algorithm penalties everyone is now starting to talk about.

4. The £100+ "all-in-one AI workspace." One tool that does emails, content, scheduling, summaries, and CRM "with AI." It does each of those 60% as well as the dedicated tool, and the AI bits are the worst part of each. Cancel. Re-evaluate after Q4 when the category settles.

The three tools worth their money in 2026 (in my audits)

None of these are categorical recommendations. They are the categories that survived the audits I ran. Substitute the closest equivalent in your country.

1. A premium model subscription. (£20/month.) Whether ChatGPT Plus, Claude Pro, or Gemini Advanced, having one premium model with high usage limits beats juggling three free tiers. Pick one. The first time you hit a free-tier wall mid-task, you will be glad of it.

2. A clipboard manager. (Free or £5/month.) Not technically AI, but the highest-leverage tool in any AI workflow. If you copy prompts a lot, a clipboard manager that lets you re-paste the last 50 things you copied saves more time than any AI feature in any app. (Mac: Maccy. Windows: built into Windows 11. Linux: CopyQ.) You do not need the £15/mo "AI clipboard" version.

3. A note-taker for calls. (£15–£25/month, OR free.) If you do more than two customer or sales calls a week, a transcription + summary tool earns its keep. Many will work fine on a free or low tier. The mistake is paying for the team plan as a solo operator.

That is it. The full "essential" stack for most one-person businesses, including the premium model, is under £35/month in 2026. Anything above that should be earning its keep on the four-question audit above.

What to spend the saved money on (sometimes)

Half of the founders who run this audit and find £400/year of cuttable spend immediately re-allocate it to other tools. That misses the point.

The point is: £400 buys you 13 Atlas Bibles, or 25 books, or two coaching sessions with someone five years ahead of you, or an entire new domain to test a side bet on. Tools have diminishing returns. Information, advice, and bets often don't.

If you must re-spend, prioritise in this order: (1) one piece of paid information per month, in a niche that compounds; (2) an actual book in your category; (3) one paid newsletter you'll read, not three you won't; (4) only after all that — a new tool.

The two renewal moments to schedule

Cuts only stick if you catch the renewal. Two reminders worth putting in your calendar today.

1. The annual review (one date, every year). Pick a single Sunday in January. Block 90 minutes. Run the four-question audit on every recurring AI subscription. By the end of the morning you will have a list of cuts and a number for the saved annual spend. Most founders I do this with come out with £200–£500 of annual savings on the first run.

The trick is to do it once a year on a fixed date, not "when I think about it" — that date never arrives. The fixed Sunday makes the audit a calendar event, not a willpower event.

2. The pre-renewal sweep (every 2 months). Open the card statement. Find any AI tool charging within the next 14 days. Run the four-question audit on just those tools. This catches the auto-renewals you would otherwise sleep through — the £40/mo tool you forgot existed, the £29/mo "promo turned regular price" you stopped using in March.

Set both as recurring calendar events. It is the highest-ROI 90 minutes a year you will spend on tooling.

What £600/year compounds to

The unsentimental case for cancellation, expressed as numbers.

£600/year saved, redirected into a low-cost index fund at a long-run 6% real return: ~£21,000 over 20 years. Redirected into one Atlas Bible per quarter (~£116/year): a steady drip of new prompt libraries that compound into your operating system. Redirected into a £600/year coaching budget: roughly six 60-minute sessions with someone five years ahead of you, which is more concentrated learning than any single tool subscription.

None of those are exhortations to do anything in particular with the saved money. The point is that the spend is not free even when the tool is "only £40/mo". The compounding is real, and it goes both ways.

The "free trial that became a habit" trap

There is a specific subspecies of AI spend that is almost always cuttable and almost always missed: the tool that began life as a 14-day free trial during a launch, became a £20/mo habit by month three, and is now an £240/year line item that nobody owns.

The diagnostic question is brutal: if this tool launched today and asked me to subscribe, would I? If the honest answer is "probably not, but I'd give the trial a go", cancel. The version of you that signed up on launch day was responding to novelty, not utility. The version of you on an ordinary Tuesday in April is the one whose vote actually counts.

One operator I ran this audit with cancelled five separate AI subscriptions in twenty minutes — none of which she could remember signing up for, all of which had been auto-renewing for at least 14 months. Total saved: £743/year. The whole audit took less time than reading this post.

What this looks like a year in

I have run a strict version of this audit on my own spend for two years. Total recurring AI tool spend in 2026: £276/year (one premium model subscription, one transcription tool on the £15/mo tier, plus Canva Pro which is half AI half not). Output: roughly the same volume of work as operators paying 4–5x that. The difference is concentration — fewer tools, used more, with prompts that have been refined over months rather than novelty subscriptions used a few times and forgotten.

The honest version of "the AI stack of a working operator in 2026" is much smaller than the marketing makes it sound. Most of what gets sold as "AI for founders" is selling the stack itself, not the work. The work needs surprisingly little of the stack to happen.

The honest disclosure

Atlas Prompts is in this category — we sell PDF "operator manuals" at £47–£60 and the starter bundle at £79. So when we say "spend on information rather than tools", we are talking our book. The reason we are still happy to write the post: even if you don't buy a bible, the audit framework above will save most readers more in cancelled subscriptions than the price of the bundle, in a single afternoon.

Run the audit. Cancel the things that fail the four questions. If you've got £47 left over and want a curated set of the prompts that genuinely replace paid tools — for newsletter ops, content production, sponsor outreach — Bible 02 (Digital Publishing Empire) is the place to start.

If you want a free version of the same idea, the free chapter walks you through one full publishing model with the prompts that, in aggregate, replace most of the £30/month "AI assistant" wrappers.

Either way: open the card statement before the next subscription auto-renews. The £600/year is not theoretical. It is in your December.