Ask, copy, paste, repeat

Watch someone use ChatGPT at work and you will usually see the same loop. They open a chat, paste something in, ask a question, read the answer, copy it, and paste it somewhere else. Then they do it again.

That loop is useful. It is also the smallest part of what these tools can do now. This article is about the bigger part, and about how to start with it.

The chat loop

In the chat loop, you do the work and the AI answers questions along the way. You find the data, bring it to the chat, check the answer, and carry the result to wherever it needs to go. Every handoff runs through your clipboard.

It feels fast because each answer arrives in seconds. But the job has as many steps as before, and most of them are still yours. When the same job comes back next week, you start from zero.

The surveys point the same way. Gallup’s surveys of US employees found that chatbots and virtual assistants are the AI tools people at work use most, and that only about one in ten employees in organizations using AI strongly agree that it has transformed how work gets done. In McKinsey’s 2025 survey, no more than 10 percent of respondents in any business function said their organization was scaling AI agents there.

Sources: Gallup, December 2025; Gallup, April 2026; McKinsey, The state of AI 2025.

What changes when AI can act

An agent is the same kind of model, set up differently. It can use your tools, it can work through several steps toward a goal, and it can be set to start on its own.

• It uses your tools. Instead of you pasting a list of invoices into a chat, the agent reads them from your accounting software. Instead of you copying a reply into your email, it saves a draft in your email app.
• It can start without you. A schedule or an event starts it: every Monday at eight, or whenever a contact form comes in. Nobody has to remember.
• It works through the steps. It checks which invoices are overdue, looks for earlier reminders, drafts new ones, and asks you to approve them, all as one job.

None of this needs a research lab. ChatGPT and Claude can both connect to email, calendars, documents, and other apps, and both can work through multi-step tasks. Some of these tools can run on a schedule or start when a new email arrives. Codex and Claude Code do the same kind of work on software. The product names change every few months. The idea stays the same.

You stay in charge

Letting AI act does not mean letting it do anything. A good setup decides up front what the agent may do on its own and what waits for a person.

My rule of thumb: reading, sorting, drafting, and summarizing can run on their own. Anything that sends, pays, deletes, or publishes waits for a person until you trust the agent with that particular job. Every run leaves a record, so you can check what happened.

Agents make mistakes. That is why the first jobs to automate are the ones where a mistake is cheap and easy to spot, and why the approval steps stay until the results are boring.

Where to start

Pick one job that comes back every week and takes you at least half an hour. Do it once more and write down every step, including each time you copy and paste. Every paste between two tools is a place where an agent could connect them directly.

Three yeses make a good first job. If one answer is no, the job may still be possible, but start somewhere easier.

If you would like help with that first job, or with the next twenty, write to me.