Build an AI workflow into your own job

You don’t need to wait for your company to roll out an AI strategy before you start using AI well in your own job. Some of the most useful setups start with one person who got tired of a weekly chore, built something that handles the first draft, and then showed their team how it works. This guide is for that person.

The approach is the same one we recommend to business owners, scaled down to a single job. Pick one task, set it up with care, test it next to the way you do it now, and keep a record of what happens.

Check what you’re allowed to use

Before you set anything up, find out what your company permits. Some businesses have an approved list of AI tools and a policy about what information can go into them. Others haven’t written anything down yet. Either way, send a quick message to your manager or IT before you paste work into anything.

If there’s no policy, stick to work that doesn’t involve customer records, financial details, passwords, personnel files, or anything covered by a contract. Plenty of useful work fits inside those limits, like drafting internal updates, cleaning up meeting notes, or organizing your own task list. Being careful here also makes it much easier to get a yes when you later ask to use AI on bigger things.

Pick a task you own

Choose a task that comes up at least weekly, that you handle from start to finish, and where a mistake would be easy to catch before it matters. Good candidates tend to look like this:

• Turning meeting notes into a summary and a list of action items.
• Writing the weekly status update your manager asks for.
• Drafting first replies to routine emails or support tickets, which you then edit and send yourself.
• Pulling numbers from a few spreadsheets into one report.
• Researching a vendor, a prospect, or a question and writing up what you found.

Avoid tasks that belong to someone else, or that would send something to a customer without you seeing it first. The first-workflow guide has a simple way to score pain and risk if you’re choosing between a few options.

Give the AI a proper place to work

Whichever tool your company allows, set up a dedicated space for the task instead of starting a new chat every time. A Grok Bot with its own instructions and files is built for this, and a project in Claude or ChatGPT can work too. What matters is that the setup remembers the job between sessions.

Put three things in it. First, written instructions that describe the task the way you’d explain it to a new coworker. Second, two or three examples of finished work you were happy with. Third, any reference files the task depends on, like a template, a style guide, or a list of terms your team uses. With those in place, you’ll stop re-explaining the job every week, and the drafts will sound more like you.

You don’t have to commit to one tool for everything. A common setup is a Grok Bot handling the recurring task, with Claude or ChatGPT brought in for long documents or careful writing. Adding Grok Bot to the AI tools you already use covers how that tends to work.

Test it next to your usual way

For a week or two, keep doing the task the way you normally would while the AI produces its own version alongside. Compare the two each time and jot down a quick note on whether the draft was usable as-is, needed edits, or wasn’t usable. Note roughly how long the review took, too.

This is extra work for a couple of weeks. It’s also what lets you say something concrete later, along the lines of “the drafts needed small edits about half the time, and the task went from an hour to about twenty minutes.” That’s far more persuasive than “it seems faster.” What to measure without fake ROI explains which numbers are worth tracking.

Keep your name on the work

Anything the AI drafts is still your work once you send it. Read every draft before it goes anywhere, and don’t let an agent send messages, update records, or make commitments on your behalf until you trust it and your company is comfortable with that. If a draft gets a fact wrong and you pass it along, the mistake is yours, so treat the review as part of the job.

It also helps to keep a short list of the things the AI regularly gets wrong, and to add them to its instructions as you go. Most setups improve noticeably over the first few weeks from that alone.

Share what works

Once the workflow has held up for a few weeks, write a short summary for your manager or team. Describe the task, how you set it up, what the AI still gets wrong, and roughly how much time it saves. Offer to share your instructions and examples so someone else can copy the setup. The format we use for showcases works well here: the problem, what you built, what it replaced, and what still needs a person.

This is often how AI adoption starts inside a company. Someone shows a working example on a real task, and other people ask for a copy. If your team decides to adopt it more widely, the 90-day adoption guide covers how to roll it out without losing what made it work. If your company wants outside help setting up agents for more of the team, a Free AI evaluation is a good place to start.