I'd rather be honest with you early than sell you a fantasy. Here's the boundary I use every single day — and why renting hits differently in 2026.
What cloud AI is genuinely brilliant for
- Writing and rewriting — emails, listings, posts, scripts, replies
- Explaining — "what does this actually mean?" in plain English
- Summarising — long threads, docs, meeting notes (check the accuracy)
- Brainstorming — names, angles, structures, first drafts
- Research reports — when Deep Research / Research mode fits the job
- Images / video — when your tier includes it
If the job is language plus structure — or media your plan covers — rented AI is often superb. No argument.
What it isn't
- Sentient. Not alive, not your mate. Software.
- Always true. Confident wrong answers happen. Check anything that matters.
- A replacement for your judgement — money, medical, legal, safety.
- Yours. They can change the models, the limits, the prices and the terms whenever they like.
The rent problem (five lines to keep)
- Free tier — enough to learn on, not enough to run a business on; caps change without warning.
- Paid stacks — £20/mo per product adds up fast (ChatGPT + Claude + Grok + …).
- Privacy — your words on their machines; fine for a pub-quiz question, wrong for client files you shouldn't be sharing.
- No ownership — your workflow breaks the day they rename a mode or kill a feature (video tools have come and gone exactly like this).
- You're on their roadmap — not building your stack.
The mental model I keep: it's a rented flat. Comfy, well kept — but the landlord can change the rules whenever they fancy. And I've had it confidently invent policies, prices and "facts" about my own projects — so treat that first draft like a keen intern, never an oracle. That one habit will save you more grief than chasing every new model name.
Privacy in practice (a preview — next lesson goes deeper)
Don't paste into cloud AI:
- Passwords, API keys, bank details
- Customer personal data you wouldn't email to a stranger
- Medical records, kids' details, anything you'd lose sleep over leaking
When local wins: sensitive drafts, internal docs, having a poke about without sending a word to someone else's cloud. We go properly deep on this in Lesson 8 — and actually running AI on your own machine is the big hands-on payoff a bit further on, once you've got the lay of the land.
Continue — open-weight models: the category that isn't "yet another ChatGPT subscription."