Back in the Jurassic era (CE 2006), my team implemented and managed an app that was, to use an oversimplified analogy, a one-to-many project management tool for retailers. It allowed a corporate employee to enter a parent project (“set up this Kid’s display”), tag the appropriate stores and departments, and the app created one child project for each of those stores.

In the stores, managers assigned each employee to departments (Kids, Cooking, etc.). Employees checked the app to see what projects were due that day for their departments, added Comments with questions, and updated project status. The corporate employee and upper management reviewed the parent project to monitor status (“complete in 50% of stores”) and respond to questions.

The app replaced a paper “planner” that was one-size-fits-all: all stores saw all projects; and all employees could see every department’s assignments. This caused confusion, wasted time, and didn’t allow remote status monitoring or two-way communication.

For the first month, we ran the app in parallel with the paper planner, while we adjusted app configurations and people got used to it. During that month, I visited stores to assess change adoption and identify necessary tweaks.

In North Carolina, one store manager said an experienced employee (“Joe”) had feedback. For context, “Joe” worked on the main floor, outside the kids department.

“With the paper planner,” Joe complained, “I can see everything. On quiet evenings – like tonight – once I’m caught up on my own work, I can flip to the next page, and knock out a few kid’s projects so they don’t have to do them all tomorrow. It made us all more efficient. Now I can only see the projects for my department and I can’t work ahead.”

The store manager said that Joe had a good point.

“What’s your sales target for today?” I asked.

“$8,763.” Our managers knew their numbers.

“It’s 8 pm and your store closes at 9, right? How close are you to your target?”

“When I checked just before you got here, down about $372.”

“How many customers are in the store?” We glanced around. They were there, hiding in the corners. “About 10? How are you going to make up that $372 in the next hour?”

“Well, it’s only about $38 per customer – we can do that. If we help every customer. By making sure they’re finding what they need, by handing them baskets if they’ve got more than 2 books, recommending related titles that they may not have noticed. Talking about how the membership program gets them a discount.”

“Is Joe the only employee on the sales floor tonight?” Yes. “What do you want Joe doing for the next hour?”

After a pause, the manager turned to Joe. “Go talk to customers. Let Kid’s worry about their projects tomorrow.”

Let me bring this out of the Jurassic.

I was reminded of this conversation when I read an article earlier this year about AI-generated employee burnout. It said that using AI agents at work causes you to feel more efficient than you are and so you keep working until you burn out. Here’s why.

When you complete work, you get a little dopamine hit 🥳: just like Joe did when he completed projects. It feels good to check something off your list. That’s why having a list is good. Every time you check something off a list you get rewarded: 🥳.

✅🥳

✅🥳

✅🥳

You get rewarded for achieving things. And you keep working, pursuing that dopamine hit 🥳 and the next 🥳. And, when you run out of opportunities for a hit, when you run out of work, you’re still craving another hit; so you look around for more work to do. You overperfect things 🥳. You “help” a colleague by doing their work 🥳.

AI automation makes it easy to check off tasks. ✅🥳

“Talking to customers” feels harder.

Customers are people and people are unpredictable. Maybe you’ll talk to them and they’ll do what you want and you’ll get that dopamine hit 🥳. But maybe they won’t do what you want. Or maybe they won’t do it NOW, it will take time to persuade them. So no dopamine hit.

Naturally, your brain wants to do the thing that will get it the sure-fire dopamine hit right now – the task ✅🥳 – over the thing that may not provide the dopamine hit.

Even if that uncertain thing has greater potential to pay off toward your goal.

Like Joe, choosing to build displays 🥳 instead of talking to customers and putting money in the till.

Now, for you, it may not be literally “talking with customers.” It may be getting your team on-board a new marketing strategy. Or coaching an employee. Or dealing with the board. Or prioritizing a list of requirements. Or translating a list of requirements into stories for engineers or designers to work on. Or checking out the competition. Prepping for a funder meeting. Figuring out a new way to do something. Making a decision. Coming up with a new strategy.

Dopamine messes with your decision making skills

A first-time author recently proudly posted his self-published book jacket on social media and asked for “our” opinion on it. 

A book jacket is not just pretty wrapping paper. It’s a marketing tool, designed to convey a story, a feeling that will make your book stand out, cause someone to be curious enough to pick it up and take it home, to feel good carrying it around where other people can see it. Professional authors fight their publishers for jacket approval rights and consider it a sign that they have “made it” when their agreement gives them the right to refuse a jacket.

I commented that this proposed jacket was too obviously AI-generated. The front cover was a generic picture that would blend in and disappear on a webpage or a bookshelf, didn’t convey anything about the content of the book, wouldn’t provoke curiosity, or make people feel good carrying it around because of what it told other people about who they are.

The author’s response was something along the lines of, “I can’t avoid using AI; it’s a better artist than me.”

“You’re better than this,” I responded. And people who understood book jackets weighed in to recommend independent artists who specialize in jacket design.

In this effort to be “efficient”, to check off the task of getting a jacket for his book ✅🥳, this very smart author, whose work I respect enormously, had overlooked the objective of the jacket: to generate sales for his book.

Finding someone to design his jacket was hard. It was “talking to customers” – he’d never done it before, so it seemed unpredictable. He’d have to find a jacket designer, maybe look at 2 or 3 to make his selection, give them a brief, correct his brief after seeing the first pass, make decisions. It was his first time, so he might make mistakes. He needed to publish his book quickly – he needed to get paid for all the time he had spent writing rather than working.

When you’re doing something for the first time, it can take a long time to get to ✅🥳. If you use AI to design a really bad jacket, it’s ✅🥳 right away.

But your book won’t sell.

Are you working for AI or is it working for you?

Another friend, a reporter, has started using AI to draft articles 🥳. He realized pretty quickly that one risk was that, if AI didn’t know the right answer, it would give a bad answer – like that annoying candidate that I interviewed for a job configuring a specific project management app. He had fooled the recruiter about his experience but, when I asked him flat out whether he had ever configured that app since it wasn’t listed on his resume, his BS answer told me he had never used it even once.

This is one reason why women are “so reluctant” to use Agentic AI: we have low tolerance for unqualified “people” confidently spouting made up sh*t when they don’t know the answer, something men are notorious for doing. One of my friends calls it “chadification (#NotAllChads)” and we know it when we see it. We hear it in meetings when the hot sh*t new hire that the boss loves because he’s so confident presents data that we know isn’t accurate. We see it when the new CIO announces he’s going to replace an operational app that we just spent two years implementing with stellar results, with a competitor’s app that we eliminated during the selection process, and which doesn’t actually do what the sales team promised him it would do. (After a year of investing time, attention, and money, the company fired the CIO and went back to the original product. Hah.) We don’t like it in colleagues and don’t like it in products.

Where was I?

Oh yes: so my friend knew he needed to check for AI hallucinations and smartly asked AI to include citations for the data it was including in his article drafts.

And then he’d use those citations to verify that AI hadn’t made sh*t up. ✅🥳

So… AI did the reporting. And the reporter had become a fact-checker.

Is that what he signed up for? I wondered.

Another Example

Another friend, an executive, is job-hunting, as we all are. They use AI to scan the job boards ✅, identify jobs that are in the neighborhood of their experience and skills ✅, automatically customize their resume to each one ✅, and submit online applications ✅. In the first month, they had sent out over 200 applications. ✅🥳

200 applications? 🥳

200 rejections.

Interviews. Zero.

“How much networking are you doing?” I ask.

Networking. Zero.

It feels like they are getting a lot done because ✅🥳✅🥳✅🥳✅🥳✅🥳. But they aren’t “talking to customers” – they’re not networking. I know, networking is hard. You have to talk to people. People are unpredictable.

If someone doesn’t give you a job or a lead on a job right away, you don’t get 🥳.

But executives rarely get jobs through online applications. Especially when every online job posting is spammed by job seekers using AI to submit applications.

Executives get jobs because someone says, “How about this person?”

That requires networking.

And now recruiters are talking about how all the resumes they receive look the same because all the applicants used AI to edit them. So job hunting is about to get disrupted yet again. At a time of record white collar unemployment. Especially for women.

About the middle of 2024, I began to see something that worried me.

I saw leaders using AI to do their employee’s work. They were choosing ✅🥳 over the work they needed to focus on: leadership.

I get it, I do. You think it will be easier and faster to just do it yourself. To show them how easy it is to just toss their awkward copy or laborious reporting into ChatGPT and come up with answers ✅🥳.

In fact, why do you need employees at all, employees who might not do what you want them to do? You’ve got AI, you can just describe what you want and it does the work ✅🥳 without you having to coach them or follow up on delegated tasks or all those things that don’t pay off immediately.

You know, the leadership things.

It’s one of the hardest things for an executive to learn. To keep out of the daily work. To soar like an eagle in the clouds and dart down to earth only when they see the shadow of a rabbit that is about to pop out of the bushes – and then to soar back up again after they’ve caught the rabbit.

They tell people how to do their jobs, instead of telling them what their job is. They say, “Let me demonstrate how to code this ✅🥳/ to write this copy ✅🥳/ to design this report ✅🥳/ edit this deck ✅🥳 – it will be faster.”

And then their employees think, “Why even bother if they’re going to redo it anyway?”

And the “leader” wakes up one day and wonders why they’re doing everyone’s work.

For my entire career, I’ve been an early adopter of business apps.

I sought them out, advocated for them, learned how to use them, trained others how to use them, figured out how to optimize them. I like new technology.

My husband, a PM, manages AI implementations all the time at work. (We both work from home, so I can hear that all his projects have been AI-related for the past 2 years.)  These AI tools – which are built by engineers; not someone with a great idea using ChatGPT – do things like screen MRIs and get a neurologist out of bed in the middle of the night if it seems like an ER patient may be having a stroke. This is important work.

His organization has an AI governance committee that makes sure that the AI tools being implemented meet security and HIPAA requirements and control for bias, something critical in the medical field, where “normal” has historically meant a white man with disastrous results for non-white men and for women, who might have different “normals.”

I’ve seen a lot of good work being done with AI in linguistics, in genetics, in pharmaceuticals, in anthropology. But again, this work is science. These tools are not being built by some guy off the street who had a great idea and used Claude to write code. The tools that do this work are built by professionals. Professionals who vet the data to prevent bias. Who test the product they built to make sure it’s not making sh*t up or sharing proprietary data with the world (like McKinsey’s AI did).

I don’t have a problem with using AI to streamline tasks – I’ve been automating workflows for decades. And testing them, going through QA and User Testing to make sure we didn’t break something. Are you doing that with your AI automations? Have you ever even written a QA or UAT script? And are you choosing what to automate and setting priorities, and not just getting caught up in automate ✅🥳, automate ✅🥳, automate ✅🥳.

Agentic AIs – the LLMs that people are using to write for them and plan vacations and analyze financial data and and and… They use the entire internet as their database and we know how much bad data is out there. The engineers who told the LLMs what to do with the prompts people give them, they work for the same people who built – and ruined – social media, filled it with false information and misogyny and discrimination and conspiracy theories that they refuse to control because it would interfere with profits.

Do you really trust these bozos not to build the same thing all over again with AI? I don’t.

Also, every bad book cover that you let AI build for you, every shortcut to building a strategy deck that “looks as good as McKinsey’s”, every AI resume or unnecessary automation has an opportunity cost: a data center. So before you “make it easier”, ask yourself: do you want that data center in your back yard? Or using up your town’s water and driving up power costs and temperatures? Or using eminent domain to take over farmland on the edge of town? Or replacing your favorite National Park or a National Forest? Because that’s the choice you’re making when you choose to use AI.

So use AI, sure.

Use it cautiously. Set a pomodoro timer that goes off every 20 minutes. When it goes off, get up, walk away, use the bathroom, get a glass of water, go for a walk, talk to a human being. Clear your head. Reset your 🥳 dopamine levels.

And make sure you’re focused on the work that helps you meet your target – not just the work that gives you 🥳. As much as it likes to pretend it can – with disastrous results that we read about in the news every day –  AI can’t really “talk to customers”. That’s the work that only you can do.

The rest of it is a trap.