AI behind a button: one-click tools for repeat work

What it is

Some tasks come back every week in the same shape: answering a common customer question, writing a status update. Each time, someone explains the task to the AI again.

"AI behind a button" means packaging that task into a small tool. The person fills in the changing part (the input), presses one button, and gets a draft back. The instructions stay fixed inside the tool, so nobody has to remember them.

Think of a coffee machine button: the recipe is set once, you press, and you still taste it before serving.

Why it matters

• Repeat work becomes faster and more consistent across the team.
• People who are not confident with AI can still use it well, because the hard part is already written.
• It shows clear, countable value: one task, done weekly, with time saved each time.

How to do it

1. Choose one task that happens at least weekly, has a clear input and a clear output, and is low risk if a draft is wrong.
2. Write down the four parts: inputs (what changes each time), fixed instructions (what never changes), output (format and length), and human check (who reviews before use).
3. Pick the simplest approved option you already have: - a saved custom assistant or project in Claude, ChatGPT, Microsoft Copilot or Gemini; - an AI add-in inside your spreadsheet or document software, where a formula or menu sends a cell's text to the AI; - a form that sends the text to an AI and returns a draft; - a no-code automation tool (one where you connect steps by clicking rather than programming).
4. Build it with made-up examples first, never real customer or personal data.
5. Test it on ten realistic, made-up inputs and compare the drafts with what a skilled colleague would write.
6. Add the human check as a visible step: the output says "Draft, to be checked by [role]".
7. Launch it with two or three colleagues, collect feedback for two weeks, then improve the instructions.

Example

Task: the team answers questions about delivery times several times a day.

Before: each person writes a reply from scratch, with different tone and sometimes the wrong policy.

After: a custom assistant called "Delivery reply" with fixed instructions:

You draft replies to customer questions about delivery.

Input: the customer's message, pasted below.

Rules:
- Use only the delivery policy in the attached file. If the answer is not in it, write: [Check with the logistics team].
- Friendly, clear, under 120 words, British English.
- Never promise a date. Use the ranges in the policy.
- Do not include any personal data beyond the customer's first name.

Output:
- Subject line
- Reply text
- One line at the end: "Draft. Checked by: [name]"

Customer message:
[paste message here]

A team member pastes the message, checks the draft against the policy and sends it themselves.

Common mistakes

• Starting big. Begin with one small weekly task where a wrong draft is easy to spot.
• Removing the human check to save time. Keep a person reviewing every output until you have strong evidence the drafts are reliable, and keep spot-checking after that.
• Using an unapproved tool. Use only approved tools, and check where the text goes.
• Letting instructions drift. Give the tool an owner and a version number.
• Not measuring. Note how long the task took before and after, so you know whether it helps.

Try it this week

1. List three tasks you did more than once last week (3 minutes).
2. Pick the one with the clearest input and output (2 minutes).
3. Write the four parts: inputs, fixed instructions, output, human check (5 minutes).
4. Save it as a custom assistant or saved prompt and test it on two made-up inputs (5 minutes).

Going further

• Spreadsheet add-ins can run the same instruction across many rows; check a sample of rows by hand, not just the first one.
• Ask the AI for structured output (for example fixed headings or separate columns) so it fits neatly into the next step.
• In no-code automation tools, add an approval step before anything is sent or saved outside your team.
• Keep a small test set of inputs and expected outputs, and rerun it whenever you change the instructions or the AI model changes.

Key words

• Input: the part that changes each time, such as a customer message.
• Fixed instructions: the rules that stay the same every time the tool runs.
• Add-in: an extra feature installed inside software you already use, such as a spreadsheet.
• No-code automation: a tool that links steps together by clicking and choosing options, without programming.
• Human check: the named step where a person reviews the output before it is used.

Examples and figures in this guide are illustrative. AI tools change often: check what your organisation's approved tools offer today. Use approved tools only, and a person checks and decides.

Prefer to print or share it? Download this guide as a PDF (243 KB), or the complete AI Field Guide (1.7 MB).