AI: £20 to £140 Overnight — And Why I’d Pay £500

An honest reckoning of what AI costs, what it’s worth, and why the bill tells a different story than the hype.

Claude Coworker is now for professionals carrying real load — no longer just for IT, developers, or marketing.

TL;DR

My AI spend went from £20 to £140 a month overnight. Most people would call that a problem. It’s the best deal I’ve ever bought. What I thought I was buying versus what I actually got. Why the headline cost is the least interesting part of the story. Why I’d pay £500.

Read this if:

•         You’re sitting on £20/month for ChatGPT and wondering whether to spend more.

•         You’ve moved to Claude and your spend has crept up and you’re nervous about it.

•         You’re trying to justify AI spend to yourself or to a board.

Start here

I went from paying £20 a month for AI to paying about £140 in a matter of weeks.

Most people would read that and see a failure — a subscription creeping up, a new dependency, a hidden cost. It’s none of those.

It’s the best deal I’ve ever bought.

This paper is the honest reckoning. What I was spending. What I’m spending now. What I got for the difference. Where the value actually lives, and why the headline number is the least interesting part of the story.

If you’re sitting where I was — £20/month on ChatGPT, wondering whether Claude is worth the switch — read this. The cost is real. The value is bigger. And the adoption curve is steeper than most people realise.

The numbers, honestly

Before: ChatGPT Pro at £20/month. Flat. Total AI spend: £20/month.

After — first three days: Claude Pro at £20/month. API usage while I worked out what I was doing — around $150 in three days. ChatGPT Pro still running as a reference at £20/month. Running total that week: ~£160, of which £150 was emergency spend.

Settled state — now: Claude Pro at £20/month. Claude 5x tier on top at $100/month (about £80). ChatGPT Pro at £20/month, kept as a reference for a while and now dropped. Total: about £140/month for Claude, at the level I now use it.

That’s a 7x increase on my old subscription. In one jump.

I’d pay more if they asked.

What I thought I was buying

Before I switched, I thought I was buying a better chatbot. A smarter conversation. Maybe cleaner writing. Possibly better analysis.

That’s what the hype sells. Smarter AI. Better answers. Faster outputs. Marginal improvements on something I already had. In that frame, £140/month is insane for a smarter chatbot.

If that’s what I’d got, I’d have cancelled in a week.

What I actually bought

Something completely different. And I didn’t know it was available at any price.

A working organisational infrastructure. Eight coworker roles, briefed, operating on their own lanes, feeding decisions back to me. Investor relations. Developer. CFO. Legal. UX. End user tester. Project controller. CEO/architect. Each with defined responsibilities, working principles, and a project index as their single source of truth.

A business inside my phone. Literally. On dispatch rides. In the car. Waiting for a coffee. I’ve taken decisions in ten-minute windows that would have taken weeks of meetings, emails, and back-and-forth if I’d been running this through humans. Not because the AI replaced the humans — because the AI produced the material the humans would otherwise have spent weeks producing, and gave it to me cleaned, structured, and ready to act on.

A second brain with discipline. The ability to think out loud with something that pushes back, holds its ground, surfaces contradictions I missed, and never forgets the rules we agreed to work by. That’s not a chatbot. That’s a colleague.

A cost compression I didn’t expect. Work that would have cost me £5,000–£20,000 to commission externally — investor document review, trademark clearance scoping, financial reconciliation, UX review, legal gap analysis — got done inside Claude in days. Not as a replacement for professional advice at the point of execution. As the preparationthat makes the professional advice far sharper and far cheaper when I do buy it.

That’s what £140/month bought. Describing it as a subscription is like describing a workshop as a toolbox. Technically accurate. Completely misses the point.

The value per pound — the number I keep coming back to

I’ve tried to sketch this honestly. Rough orders of magnitude, not accountancy.

£140/month is £1,680/year.

Against that, in the first six months of using Claude seriously, I’ve produced — or had Claude produce under my direction — work that would have cost me, conservatively:

•         Investor pack review and reconciliation: £3,000–£8,000 if I’d commissioned a fractional CFO or investor relations consultant.

•         Legal gap analysis and trademark scoping: £2,000–£5,000 before a solicitor even picks up the phone.

•         UX and flow analysis across a 71-screen prototype: £5,000–£15,000 with a UX consultancy.

•         Project management and document control: £20,000+/year for a part-time project manager.

•         Strategic thinking partner / sparring on decisions: incalculable, because I’d never have afforded it as a line item.

Total floor: somewhere around £30,000 of work, produced on a £1,680 subscription.

Ratio: around 18:1.

I’m not telling you that number to sell you anything. I’m telling you it’s the reason I’d pay £500/month without flinching. Because the ratio would still be five or six to one at that price, and it would still be the cheapest professional infrastructure I’ve ever had access to.

Where the value actually lives

It doesn’t live in the AI being smart. Plenty of AIs are smart now. The value lives in three places most people miss.

It lives in the setup. A properly set-up Claude workspace — preferences right, coworkers briefed, project index maintained, working principles embedded — produces output at a level a badly-set-up one simply can’t reach. Same tool. Completely different results. That’s why Papers 1 and 2 in this series exist. Without them, the £140 is wasted.

It lives in the operator. The AI is only as good as the person using it. You get paid for what you know how to ask, how to test the answer, how to route the work, and how to catch the drift. Most people plateau fast because they don’t develop these skills. The ones who do, compound.

It lives in the cadence. Daily use, threaded properly, with the right discipline, builds a working relationship that gets sharper every week. The value isn’t in any single conversation. It’s in the hundredth conversation, where the infrastructure you’ve built is doing work in the background you barely notice.

The tool is a commodity. The leverage is everything around it.

What I pay for, specifically

Broken down so you can judge for yourself.

Claude Pro (£20/month): The core. Fine for light use. You’ll hit the limits fast if you’re doing serious project work.

Claude 5x tier ($100/month, ~£80): Stops me hitting the wall. Lets me work in flow without rationing. This is the single biggest value unlock — not because the AI gets smarter at higher tiers, but because the friction disappears. No mental overhead of “should I be saving this for later.” Just work.

Claude Cowork: Where the coworkers actually live. Included at the tiers I’m paying. The environment that makes multi-coworker project work possible.

If I were starting from scratch today, knowing what I know now, I’d go straight to the 5x tier. Pro alone is training wheels. Useful to learn the shape of the tool. Not enough for real operational use.

The adoption curve — what I keep telling people

Here’s what I think is coming, held as a view with conviction but no certainty.

In a year, serious operators in every knowledge-work field will be running some version of what I’ve just described. Coworkers, project indexes, preferences, working principles. The early adopters will be obvious — the output gap between them and the rest of the market will be wide enough to see.

In three years, this will be table stakes. Not having an AI workspace set up properly will look the way not having email looked in 2005 — technically possible, professionally embarrassing. The tooling will be better, the setup will be easier, and the price will probably come down. But the operators who started now will be three years ahead, and that’s a lead that doesn’t close quickly.

The window isn’t “should I adopt.” It’s “how far ahead do I want to be when everyone else catches up.”

At £140/month, the question isn’t whether the bill is worth it. It’s whether you can afford not to be building this infrastructure while it’s still an edge.

What I don’t pay for

Fair to name these too, because they’re where people think the value lives and it doesn’t.

I don’t pay for faster writing. Nice to have. Not the point.

I don’t pay for saved time on emails. Marginal. Honestly trivial.

I don’t pay for “ideas.” I have plenty. The AI’s ideas are mostly my ideas, reflected back with structure. Useful, but not the bill-shifting value.

I don’t pay for the novelty. The novelty wore off in a week.

I pay for the infrastructure. Everything else is garnish.

One last thing

The headline number is £140/month and most people will stop reading at the headline.

The people who keep reading are the ones asking a different question — what does it actually produce, and what does that compare to? Those are the ones who’ll be running proper AI infrastructure in a year, and leading their markets in three.

The bill went up. The value went up further. That’s the whole story.

Pay attention to the ratio. Not the price.

Before you do anything else

Grab a pen. Or a blank page.

Do the maths on your own stack. Two columns.

•         Column A — What you pay for AI now. All of it. Subscriptions, API, the lot. Per month.

•         Column B — What that AI actually produces in a month. Not in hours saved. In work output. What would you pay a human to produce the same work? Be honest. Include the things you couldn’t have afforded before — the thinking partner, the reviewer, the structured briefs. Count them.

Work out the ratio. If Column B divided by Column A is less than 3, you’re either under-using the AI or the setup is wrong — see Papers 1 and 2. If it’s more than 10, you’re in the zone most people will be in three years from now.

Then ask yourself: what would I pay for this if the bill doubled?

If the answer is still yes, you’ve bought the right tool. If the answer is no, go back and look at what you’re actually using it for. Chances are, it’s not infrastructure yet — it’s still a chatbot. That’s the gap this series is written to close.

It’s not a subscription. It’s infrastructure.

Concept, chain of thought, and content: Paul Roebuck | Words: Claude AI Opus 4.7 | images: ChatGPT5.5 | Paper 9 | 2026 | paulsroebuck@gmail.com | https://paulroebuck.co.uk/contact | https://paulroebuck.co.uk/blogai