AI Guides › Playbooks
By Nigel Guy · 5 min read
Most people who get excited about an AI tool go straight to building an app. The idea feels concrete once it has a logo and a login screen, so it feels like progress, and it is also the most expensive way to find out whether anyone, including you, would use the thing. A recent headline in circulation claimed Google cancelled its own AI Studio app after 800,000 people pre-ordered it. We could not verify that: AI Studio is a web tool you open in a browser, and we found no official source describing pre-orders or a cancelled app. Treat the story as unconfirmed. The lesson people draw from it is still worth having, and it does not need the story to be true.
The rule: do not build an app until the same job has failed on a saved prompt and then on a shareable page.
You climb one rung at a time and only move up when the rung below has visibly broken. Each rung has an exit test, written before you start.
| Rung | What it is | Move up when |
|---|---|---|
| 1. Saved prompt | A reusable prompt in a chat or project, with your inputs in placeholders | You have run it at least five times on real inputs and the copy-paste is the bottleneck |
| 2. Shareable page | A single-page tool or document made as an artifact, or a spreadsheet | Other people need to use it without you, or you need stored data |
| 3. Real app | A deployed, maintained application | Rung 2 has hit a limit you can name: logins, payments, scale, or integrations |
Most "app ideas" are a prompt with a form around it. Write the prompt first and run it on real material. This one works for any idea:
You are a pragmatic product advisor helping me decide whether a small tool is worth building.
My idea: [IDEA_IN_ONE_SENTENCE]
Who would use it: [PERSON_AND_SITUATION]
What they do today instead: [CURRENT_WORKAROUND]
Sample real input I can give it: [EXAMPLE_INPUT]
Do this in order:
1. Restate the job the tool does in one sentence, in the form "takes X and produces Y".
2. Produce the output for my sample input, exactly as the tool would.
3. List what a form, login or database would add beyond this prompt. Be specific.
4. Say which rung of this ladder I need: saved prompt, shareable page, or real app.
Format: four short numbered sections, no more than 250 words total.
Rules: if any bracketed input is missing, ask me for it instead of guessing. Do not invent users, numbers or market sizes.
Before answering, check that step 4 follows from step 3 and say so in one line.
Fill in the five bracketed fields. If step 3 comes back with nothing concrete, you have your answer: stay on rung 1.
Claude's artifacts are standalone pieces such as small tools, dashboards or documents that open beside the conversation. They start private and you can share them by link. At time of writing, Anthropic's help centre lists artifact creation in chat on all plans, with templates, connected apps and data storage on paid plans (Pro, Max, Team, Enterprise), and cloud code execution and file creation must be switched on under Settings > Capabilities. Check the current plan table before you rely on any of this, because it moves. Prices change too, so check the £ price at checkout rather than trusting a figure in a guide.
Google's AI Studio has a Build mode that sits at roughly the same rung: you describe an app, it generates a live preview, and you can download it as a ZIP or sync it with GitHub. Per Google's documentation, deploying to Google Cloud Run is a separate step with its own usage-based costs. That is the point where a prototype starts becoming a thing you maintain.
Only climb here with a written reason. Acceptable reasons: you need user accounts, you take payments, you need data that must survive and be backed up, or you need it to connect to systems that a page cannot reach. "It would look more professional" is not on the list.
Imagine a freelance bookkeeper who wants an app that turns messy client emails into a monthly "what I still need from you" list. On rung 1 she saves a prompt, pastes in five real emails over a fortnight, and gets usable lists. The bottleneck is now that her clients cannot paste emails themselves. That justifies rung 2: a shareable page with a text box, which she tests with two clients. Neither needs a login. The ladder stops there. An app would have added hosting, a privacy policy question and ongoing upkeep for no extra value.