5 ways to actually use AI chat at work (that aren't meeting minutes)

None of these are about writing faster. They’re about thinking better, in the fifteen minutes you actually have between meetings. Here’s how to use each one on purpose, not just once you stumble into it.

1. Dump your head before dinner, but make it a plan

The habit is simple: end of day, everything that’s still rattling around gets dumped into the chat. Here’s the part worth knowing so you get the actual benefit, not just a shorter to-do list.

Psychologists E.J. Masicampo and Roy Baumeister found that unfinished tasks create intrusive thoughts that interfere with focus on whatever comes next, the classic “why can’t I stop thinking about work at dinner” problem. Their follow-up finding is the useful part: simply writing the task down doesn’t clear it. Writing a specific plan for finishing it does. Once people had a concrete next step on paper, the intrusive thoughts stopped. Source: Psychology Today, “How the Little-Known Zeigarnik Effect Impacts Everyone Daily”

The upgrade: don’t just list what’s unfinished. For each thing, add one line: “next step is X.” That’s the difference between a dump that still follows you to dinner and one that actually lets you put your phone down.

2. Sort the mess when you’ve got ten ideas and no order

When everything feels equally urgent, nothing gets started. Paste the whole tangle in and ask it to group what’s related, flag what’s actually time-sensitive, and hand you back three things to look at first. You’re not outsourcing the decision, you’re just getting the sorting done faster than you’d do it staring at a blank page.

3. Pressure-test a decision before you commit to it

This is the one that changes the most, once you actually use it this way.

Research on AI-assisted decision-making shows that an AI “devil’s advocate,” something that actively questions your reasoning instead of just agreeing with it, leads to higher-quality decisions and a better experience for the person deciding. The mechanism is the same one behind minority-dissent research in psychology: disagreement forces you to actually examine your reasoning instead of coasting on it, even when the pushback turns out to be wrong. Source: ACM, “Enhancing AI-Assisted Group Decision Making through LLM-Powered Devil’s Advocate”

How to actually do it: don’t ask “is this a good idea.” Ask “what’s the strongest case against this” or “what am I not accounting for.” The default answer from most AI chat is agreeable. You have to ask for the friction on purpose.

4. Practice the hard conversation before you have it

Type out roughly what you need to say, and ask it to play the other person, realistically, including the pushback you’re actually dreading. Then respond like you would in the room. You’re not looking for a script to read off. You’re looking for the version of the conversation where you’ve already heard the hard part once, so it doesn’t catch you off guard live.

5. Ask it to challenge your assumptions

Different from pressure-testing a single decision. This is for the stuff you haven’t questioned in a while: how you’re running a meeting, how you’re structuring a project, how you’ve always done something. Ask directly: “what am I assuming here that might not be true.” The value isn’t that it’s always right. It’s that it makes you say the assumption out loud, which is usually the first time you’ve actually looked at it.

A copy-paste prompt to try any of these tonight

Here's what's still in my head from today: [paste the list]. For each one, help me write
one concrete next step, not just a restatement of the task. Then ask me one question that
challenges an assumption I'm making about my week.

If this is the kind of thinking you want to build a habit around

This is exactly the muscle my course, AI, Finally, is built to teach: using AI chat as a reasoning partner at work, not just a faster typist. It’s linked in my Stan store if you want the fuller version of this.

Try it this week

Pick one of the five. Use it once, on purpose, this week. Not all five at once, just one, so it actually sticks.