You should be using an AI chatbot at work

Today I'm writing about what everyone else is writing about: artificial intelligence. Very few opinions from me here, however, and more about how to use it. Because you should be using it.

That's not my opinion. That's the opinion of Ethan Mollick, a recent guest on the Ezra Klein Show, for the episode "How Should I Be Using AI Right Now?"

"Bring it to every table," says Mollick, meaning you should be using it for every task—writing emails, planning meetings, brainstorming ... save for anything to do with PHI. Wait for an EHR integration to do that.

You're bound to get something useful out of your experimentation. More importantly: you'll have started using AI. That's important because turning to AI to support your work tasks feels unnatural at first. And it slows work down. But in using the AI chatbot you'll discover its abilities and where it can best support your work, because as Mollick says, “the key is to use it in an area where you have expertise so you can understand what it is good or bad at.”

He also has observed a ten hour rule: to reach proficiency in interacting with an AI chatbot requires ten hours of use. This moves you past the just poking around phase and toward skill development. Using an AI chatbot is a new skill to learn, and if you haven't yet, you're going to be pleasantly surprised how good it is.

According to Mollick, AI is better than the average person at a random task, noting: “Part of the reason why I advocate for people to use it in their jobs is it isn’t going to outcompete you at what you’re best at." Not yet, anyway. And likely not ever.

One mistake I made early in my use of the AI chatbots was treating the AI like a computer. Hmph. Silly me. I made this decision with intention based on the fact that ... well, it’s a computer. However, Mollick advises to approach using the AI not as a software tool like Google, but like you would in building a relationship with another person because it was designed to interact like humans in conversation. I found this change greatly reduced the friction I felt when prompting that chatbots.

Okay, prompts! This is important, at least for now, as Mollick believes the AI tools are going to rapidly improve and prompt quality won't matter nearly as much as it does right now. Good prompts are more likely to lead to good output. The big problem in AI models at the moment is that with an average prompt you'll get an average answer back—which to be clear is often amazing since, you know, we're interacting with a computer! But we don't want average! We want useful!

So Mollick advises three methods of prompting the AI chatbot: Chain-of-thought, personas, and few-shot.

Chain-of-Thought

• Instruct the AI step-by-step, one instruction at a time.

  • Prompt 1: Write an outline for a PowerPoint presentation about improving meeting management

  • Prompt 2: Write the headline for each slide of the PowerPoint presentation

  • Prompt 3: Write appropriate bullet points for each slide of the PowerPoint presentation

  • Prompt 4: Review the PowerPoint presentation and make edits to improve quality

Personas

• Personas: Give the AI a persona and context of what you're seeking.

  • "You're a McKinsey consultant, I've pasted an email I'm sending to the leadership team, please provide advice for improving the clarity of my message."

  • "You're an executive coach. I'm having challenges with a colleague who seems to be undermining my work, how do you recommend I proceed?"

  • "You're Taylor Swift's PR team, what advice do you have for launching this new internal marketing campaign for getting our customer service employees to think about the experience of customers they're helping on the phone?"

Few-Shot

• Provide some examples ("shots") to the AI, and then provide the prompt. (Everything below is your prompt—you'll be amazed at how good the chatbot is at figuring it all out!)

  • Example 1: [paste email copy] Here's an example of a project update email I wrote.

  • Example 2: [paste email copy] Here's an example of another project update email I wrote.

  • Prompt: Write me a project update email based on the meeting transcript pasted below.

Chances are you’ll get even better output through a continued back-and-forth with the AI, no matter which type of prompt you use. Tell it what you like, tell it what isn’t working, iterate your prompt. Keep going. The good stuff comes as you continue the conversation.

I'm using Anthropic's Claude (Mollick says it's best for writing), though I started with Chat GPT (according to Mollick it's best for math and coding; it had a major update this week ... and some advanced features are now available in the free version), and the other to consider (though there are many, many, many available ... ) is Google's Gemini (Mollick: "it really, really wants to help").

My advice is to pay the monthly fee to get access to the latest model of whatever tool you choose. There's a huge difference in quality between the free version and the version behind the paywall.

To summarize:

1. Sign up for an account

2. Bring your AI work companion to every table

3. Prompt it like you would a conversation with a friend

Around The Water Cooler ⛲

Ethan Mollick’s AI newsletter. It’s great!

Evening Relaxing: Shows I’m excited are back for their third seasons: Welcome to Wrexham, Clarkson’s Farm, and Hacks

Anne Helen Petersen is at least in part responsible for elevating the topic of burnout into popular awareness, her book is a good intro, and also: How I've Changed My Thinking About Burnout

How To Work is healthcare-focused work design inspiration (from the experts!) to nudge your perspectives and practices into better alignment with the world of work as it is, and away from what it was. Read my take on the tension we’re working through.

Are you interested in a specific topic? Simply tap “Reply” to send me a question or comment.