Chapter 16 · Part Four: The Machine That Makes Average
The best AI adoption story I know didn’t come from marketing. It came from a CISO.
Benjamin Dulieu is the CIO and CISO at Duck Creek Technologies. He told this story on our podcast, Do Human Work, and he’s been generous enough to let me retell it.
He was a believer in AI early. And he understood something most people pushing AI inside a company won’t say out loud. A lot of what he was about to pitch to his team, if you squinted, sounded like telling them to replace themselves.
But he had a conviction. If he could get his small team out of the repeatable work, they’d do the strategic work they had never had the hours for.
The problem with a conviction is that going from “I believe this” to anybody changing what they do on a Tuesday is enormously hard. “You should use AI more” just sounds like more work, from somebody who isn’t doing it.
So he called one of his GRC managers. Smart, pragmatic, hungry.
And he asked her one question.
Do you want to get promoted this year?
Yeah.
Do you want to go above your max bonus?
Yeah.
Then take the security questionnaires, the ones the team spends hundreds of hours a month answering, and fully automate them.
If you’ve never had the pleasure: the work has to get done, but it doesn’t need a human at a keyboard every single time. The answers already exist somewhere. Somebody just has to assemble and check them.
That was a Thursday night. The next Wednesday she walked into his office. The first thing she said was thank you. She’d like to believe she would have gotten there on her own. She just needed the opportunity.
Five business days, and she had version one. It’s an agent running at Duck Creek today, supporting the pre-sales team, and he says it saves them hundreds of thousands of dollars a year. He also says it did something more important than that: it showed the rest of the company what was possible, and people who see it become believers.
Why the mechanic works
Look at what he actually did, because it’s a better change-management design than most consultancies sell.
He asked a question nobody says no to. Not “will you adopt AI?” which invites a debate. “Do you want to get promoted?” There’s no debate.
He picked the right person and the right task. Somebody hungry, and a job everybody in security knows is necessary, repetitive, and miserable. Not a moonshot. One basic use case, chosen because success would be obvious to everyone who’d suffered through it.
He tied it to something real. Not a hackathon prize. Her promotion and her bonus.
He gave her permission. This is the part I’d underline, and it’s in her own words. She didn’t need training, a tool budget, or a working group. She needed somebody senior to say: this is your job now, go. “I just needed the opportunity” is what most people on most teams would say if you asked them.
None of that costs money. All of it is available to anybody who manages anybody, this quarter.
The suspicion
Here’s why the story lands so hard with people, and why I think everyone reading this can feel all three parts of it.
The first part is the uncomfortable suspicion that the tasks are your value. That if a model can do the security questionnaire, maybe there’s less reason for you.
I had a version of this conversation early at 7AI, interviewing a candidate for a sales leadership role. He looked at what our product did and worried that it would be hard to sell, because it was going to take away jobs in security.
So I asked him: if I automated away your need to put notes into Salesforce, would you be out of a job?
“Are you kidding me? Sales is way more than that.”
Bingo.
Nobody believes their own job is the data entry. Everybody suspects that about everybody else’s.
The second part is going looking for your security questionnaire. And the third part, which is the only one that matters, is what becomes possible once you stop doing it.
Your value was never the tasks.
The ten hours
Here’s a question I ask people in marketing and in security, and I want you to actually answer it before you keep reading.
If I handed you back a quarter of your week, starting Monday, ten hours, what would you do with it?
I’ve asked a lot of people. Nobody has ever said they’d make more slides. Nobody says update the CRM, or write another status report, or sit in another meeting where we align on alignment.
They say they’d talk to customers. They’d think. They’d build the thing they’ve been putting off for a year. They’d fix the process everybody complains about.
When we asked Benjamin, his answer was the best one I’ve heard. Time with his family, and taking apart the inefficient things everybody in security has stared at for their whole career. He boiled it down to one line: making memories and breaking systems.
You already know what your best work is. You’ve known the whole time. The only question is what you’re going to stop doing on autopilot so you can go do it.
The Task Inventory
The tool for this chapter. An hour, alone, with your calendar from the last two weeks.
Step 1. List the tasks. Not your responsibilities. The actual tasks. Everything you did that took more than fifteen minutes, written as a verb and an object. “Built the weekly pipeline report.” “Answered the same positioning question from a new rep.” “Reformatted the deck for the partner.” Aim for thirty.
Step 2. Run four tests on each.
- Does it repeat? Same shape, at least monthly.
- Do the inputs already exist? The answers live somewhere: a document, a system, a past version.
- Can you say what good looks like? If you can’t describe a correct output, a model can’t produce one, and neither can a new hire.
- Does it need your judgment every time, or only at the end? Most tasks need a human to check, not a human to assemble.
A task that passes all four is a candidate. A task that fails the fourth test is probably your actual job, and you should be suspicious of anybody, including yourself, who wants to automate it.
Step 3. Score the candidates. Hours per month, multiplied by dread on a scale of one to five. Dread matters. The task you hate is the one you do worst, put off longest, and will be happiest to stop doing, which makes it the one you’ll actually follow through on.
Step 4. Pick the top one. One. Not a roadmap. Give it a deadline and the permission to drop everything else, the way Benjamin did.
Step 5. Write down the ten hours. Before you automate anything, write down what you’ll do with the time it frees. If you don’t, the time will be filled by the inbox within a fortnight, and you’ll have built a crutch.
The Comfort Check
Chapter 16: your own job
1. The comfortable assumption. My value is the work I do every week.
2. The true objective. Spend my hours on the work only I can do, and know what that work is.
3. The redesigned approach. Inventory the tasks, automate the one with the most hours times dread, and decide in advance where the freed time goes.