How to Start Using AI at Work

Start with one repeating task, not a tour of tools

People start using AI at work by collecting logins. They try ChatGPT, a copilot, a meeting notetaker, and a browser sidebar in the same week. They still cannot point to a task that got faster and safer. Starting well is narrower. Choose one repeating artefact you already own, such as a Tuesday status note. Confirm the tool is approved for that data class. Write a brief before you open it. Generate once. Spend the recovered minutes checking facts and tone. Send only what you would have sent if you had typed it. That loop, five times, is a start. A tour of features is entertainment. Entertainment does not compound.

The usual stall is waiting for a training day, or waiting for the perfect use case. The perfect use case is the work that already has an owner, a reader, and a source. If you do not own the close, do not start on the close. If you own a weekly agenda, start there. Guardrails belong in the first hour, not after a scare. You need a rule for what does not get pasted, a named reviewer if the artefact leaves the team, and a place to save the brief that worked. Without those, you will either leak something or abandon the tool when the first fluent error appears. Both outcomes are common. Both are avoidable.

This essay is a first-week method, not a catalogue of products. It assumes your organisation has at least one approved place to type. If it does not, the start is a conversation with security, not a consumer account on your phone. A named human still owns the send. AI at work begins when that human has a boring loop, not when they can name six models. Skip the tour. Do the Tuesday note. Write down what the model got wrong. That note is the beginning of a team habit. Start there.

Task, tool, brief, review: four pieces, one loop

The task must repeat, or you will not learn. A weekly status, a meeting agenda, or a short internal explainer is a good one. The tool must be approved for the data you will actually use. Do not start in a personal account because the work laptop blocks the site. That is how customer text ends up in the wrong place. The brief must exist in writing: reader, ask, facts, bans, output shape. The review must be a step with time on the calendar, not a hope that you will notice errors. If any of the four is missing, you do not have a start. You have an experiment that will not survive a busy Thursday.

Measure the loop by time-to-trusted-send, not by words generated. If the draft takes longer to clean than to write, change the brief or stop. If the draft is faster and still true, keep it. After five cycles, save the brief in a shared note with an owner. That is adoption. Adoption is not a slide that says we are exploring AI. Exploring is what people say when there is no artefact. You want a Tuesday note that used to take forty minutes and now takes twenty-five with a better check. That sentence is the whole frame. Everything else can wait until the loop is boring.

Five cycles on the same artefact

Day one: write the brief and the bans, including what you will never paste. Generate the Tuesday note once. Check every name, date, and number against the file. Send. Write two lines: what was wrong, what you changed. Day two through Thursday: same brief, same artefact, same review. Change only one instruction if the tone or length is wrong. Do not add a second tool. Do not add a second artefact. Friday: decide whether the brief earned a place in a shared note. If it did not, stop or pick a simpler task. Starting is a decision to continue or to halt with evidence, not a mood that you should use AI more.

Invite one colleague to run the same loop on their own repeating task, not on yours. Parallel practice beats a demo in a meeting. Keep the guardrails identical: approved tool, no secrets, named sender. If your manager asks what you are doing with AI, show the brief, the log, and a before-and-after send. Do not show a clever chat. Clever chats do not survive holidays. Briefs with owners do. When the first week is done, you may add a second artefact. You still do not need a tour of plugins. Plugins are how the loop becomes a hobby. The hobby is how people paste the wrong file.

Role: You are helping me run a first-week workplace AI loop I will own.
Task: Draft the repeating artefact I describe from the facts I paste.
Context: I will name the reader, the day this goes out, and the approved tool.
Constraints:
- Use only facts in the paste. If something is missing, write [missing].
- Do not add tools, plugins, or extra artefacts to this task.
- Do not invent dates, owners, or status.
Output: The artefact, then a short list of claims I must check against my file.
Quality checks: What would go wrong if I sent this without opening the source.

Run that prompt on the same artefact five times this week. Do not switch products when the second draft is awkward. Change one line of the brief. Check the file every time. On Friday, either save the brief with your name on it or drop the task and write down why. That written decision is the start. A collection of unused logins is not a start, and it will not help the next person who asks how you use AI at work.

Tours, personal accounts, and starting on work you do not own

A tool tour creates the feeling of adoption without an artefact. You will remember features and forget the review step. Personal accounts create a second risk: the work is now in a place your administrator cannot see, on terms you did not negotiate. Starting on work you do not own, such as rewriting someone else's board paper, creates a third: you cannot check the numbers, and you still might send the tone. Stay inside a task you can verify. Stay inside the approved tool. Stay inside a brief you can show. If security has not named a tool, do not invent a start-up exception. The first week is where habits lock. Lock a boring safe loop, not an exciting leak.

Watch the urge to skip the log because the first send looked fine. The miss arrives on the fourth cycle, when you are faster and less careful. The log is how you notice that dates drift or that the model keeps adding a thanks for a concession that did not happen. Also watch shadow training. A lunch-and-learn that shows twenty prompts and no review step will undo the week. If you are asked to present, present the four pieces and the two-line log. That is how a start becomes a team method instead of a private amusement that dies in week two.

Related reading on StudyGrid

Read next: Practice and Reference 30-Day AI at Work Plan How to Use ChatGPT at Work. Those essays sit beside this one. Use them when you need the neighbouring skill, not as a substitute for the check you still have to make.

What to do this week

Pick the artefact you will send next Tuesday. Confirm the approved tool. Write a brief and a paste ban. Generate once. Check the file. Send. Repeat on the same artefact through Friday. Save the brief if it earned the time, or stop with a written reason. Skip every other product until that loop is boring. Keep the two-line log. That is how you start using AI at work without a tour.