Automate, Delegate, Do: A Framework for Discernment in the Age of AI
- FEATURED ARTICLE
Automate, Delegate, Do: A Framework for Discernment in the Age of AI
Matt Bradley (Partnership Manager, WhyFire)
Share
Listen to the audio article.
Computer Games and Funny Emails
In one of my favorite episodes of The Office, Michael Scott finds himself in a college lecture hall, explaining the finer points of business to a room full of future MBAs.
The setup is simple: Michael’s young employee, Ryan, is in business school, and bringing his boss to class automatically bumps his grade up a full letter. So Ryan invites Michael, telling the camera he’d be stupid not to—right?
Michael, of course, is thrilled. Although he was actually invited as a simple guest speaker, he decides he’s being honored as a visiting professor, and he shows up ready to inspire.
The lecture goes about how you’d expect. Michael opens by throwing out candy, making cheesy puns, and ripping pages out of a student’s textbook—proof, he explains, that you can’t learn business from books. When someone finally steers him toward a question-and-answer session, a student raises his hand and asks Michael the following question: “[A]s a company that primarily distributes paper, how have you adapted your business model to function in an increasingly paperless world?”
Michael doesn’t miss a beat: “We can’t overestimate the value of computers,” he says. “Yes, they are great for playing games and forwarding funny e-mails, but real business is done on paper, okay? Write that down.”
And then all the students start typing like mad, just as they’ve been told to do.
This is one of the countless scenes in the show that reveals its wit and wisdom—because even though it’s downright hilarious, it also reveals a deeper truth: Michael has no idea how to use the most powerful tool in his office.
Consider his position: As the regional manager of a paper company, Michael lives in a world of quotes, clients, invoices, and inventory—exactly the kind of work a computer is built to handle. Used well, the machine on his desk should help him do more work—better and faster.
But used his way, it’s a toy for playing pointless games and forwarding funny e-mails. It’s a distraction that leaves him doing less work—worse and slower.
Of course, the problem here isn’t the tool itself; instead, it’s how Michael uses it.
And that’s the way it works with virtually every tool, from the oldest to the newest.
Think about it: If you use a hammer to hit a nail on the head, that’s a good use of the tool. But if you use it to hit yourself on the head, that’s a bad use of the tool.
Simply put, using any tool well comes down to discernment.
And this is especially true of the newest, most powerful tool on the planet: AI.
After all, discernment is what tells you which tasks AI should run on its own, which tasks AI should draft for your review, and which tasks belong to you alone (no matter how good the models get).
But as Michael proved in front of that class, discernment doesn’t come with any given tool. Instead, it’s something you have to develop on your own.
That’s why, over the last few years, I’ve developed a system for discerning when and how to pass off work to AI, one that’s built on three small words: automate, delegate, do. More specifically, if AI can run a task on its own (or build something that can), then I automate it. If AI can draft a task for my review, then I delegate it. And if a task belongs to me alone, then I do it.
To be clear, there’s nothing revolutionary about this system. It’s just a practical way of practicing discernment that’s helped me do more work—better and faster.
With all that in mind, let’s take a closer look at each of these groups now, then put them to work inside a hearth business like yours.
Simply put, using any tool well comes down to discernment.
Automate
The first group holds the tasks AI can take off your plate entirely—either by doing them itself or, better yet, by building a simple system that does them without it.
When I talk about automating a task, I mean taking the repetitive, rules-based, low-stakes work off your hands for good—usually by having AI build a simple system that runs end-to-end. The setup may take some time, but once the system is built, nobody needs to be in the loop: not you and, most of the time, not even AI.
So how do you know whether a task belongs in this first group? Ask yourself three questions.
- Is the task repetitive (basically the same job, again and again)?
- Is it rules-based (so predictable that you can define exactly when it fires)?
- And is it low-stakes (meaning no one gets hurt if AI gets it wrong)?
If the answer to all three questions is “yes,” then that’s a task worth automating.
This is the group that author Joanna Maciejewska captured in a line I’ve shared in this publication before: “I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes.” Notice what her line assumes: The tool is neither the hero nor the villain of the story. It’s the servant of a well-ordered life—and the laundry goes into the machine so that the art can stay with the human.
So what does the laundry look like in a hearth shop? Consider the follow-up work sitting inside your sales pipeline right now. Every week, customers visit your showroom, receive quotes, and book consultations—and every one of those moments deserves a timely follow-up message. That work is repetitive because the messages take the same shape every time. That work is rules-based because you can define exactly when each message should fire. And that work is low-stakes because the worst thing a misfire can produce is an extra friendly email. Three questions, three yeses—so this is a task you should automate.
If you’d like to do exactly that, copy and paste the following prompt into your favorite AI tool, filling in the templated parts with your actual information:
“I own a local hearth shop, and I want a simple, automated follow-up system for my sales pipeline. I usually use [program name] to create spreadsheets, and I send emails through [program name]. Help me build a spreadsheet that tracks every lead—including the customer’s name, contact info, sales pipeline stage, and date of last contact—and automatically sends the right follow-up email whenever a customer sits in one stage for more than three days. Walk me through creating the spreadsheet and setting up the automation one step at a time.”
Take an afternoon to build that follow-up engine with AI, and it will quietly repay you for years to come. No lead will fall through the cracks while you’re out running installs, and no customers will wonder whether your shop forgot about them—because the follow-ups will go out morning after morning, whether you’re at your desk or not.
In other words, give the laundry to the machine, and give the art your full attention.
This article is brought to you by Associated Energy Systems.
Delegate
Of course, a follow-up engine like that raises an important question: Who writes the emails it sends?
That question matters more than it might seem. After all, the schedule may be mundane, but the words aren’t. Every one of those emails carries your company’s name into a customer’s inbox. And work that carries your name calls for a different arrangement with AI, which brings us to the second group in the system: the tasks AI should draft for your review.
When I talk about delegating a task, I mean giving AI the nuanced, judgment-heavy, mid-stakes work it can draft but shouldn’t ship. Drafts will usually need revision, so no output goes out without human review and approval.
The test for this group works the same way as the first. Ask yourself three questions, and look for three yeses:
- Is the task nuanced (as in, does it require your vision or voice)?
- Is it judgment-heavy (would you want to read the output before it goes out)?
- And is it mid-stakes (meaning a mistake would cause real consequences, but not lasting ones)?
The emails inside your follow-up engine answer “yes” on every count. They’re nuanced because customers can smell a canned form letter from across the room—and they want to hear from a real person instead. They’re judgment-heavy because you’d never let a stranger send email on your company’s behalf, and an unreviewed draft is exactly that. And they’re mid-stakes because a clumsy line might cost you a sale, but it won’t cost you your reputation. So don’t automate the writing of those emails—delegate it, with a prompt like this one:
“Here are three examples of emails I’ve written to my hearth customers, so you can learn my voice: [Paste two or three of your best here]. Study them carefully, then draft three follow-up emails my automated system will send: one checking in three days after we’ve sent an online estimate, one thanking a customer for visiting our showroom and asking for an in-home appointment, and one asking for a deposit after an in-home appointment has been completed. Match my voice, keep each email under 150 words, and leave a [bracketed placeholder] anywhere a detail like a name, product, or price belongs. I’ll review and edit every draft before anything gets sent.”
I want to stress the last line of that prompt because the review is where delegation succeeds or fails. On a recent episode of the Lex Fridman Podcast, software entrepreneur Peter Steinberger—a man who’s spent his entire career building software—explained why so many people abandon AI after one disappointing attempt. Using AI well, he argued, is “a skill that you have to learn like any other skill.” In the same way you have to practice playing the piano before you can make beautiful music with it, he explained, you have to practice using AI before you can make anything valuable with it. Using AI “needs a different level of thinking,” he concluded. “You have to learn the language of [AI agents] a little bit, understand where they are good and where they need help.”
His point is worth pausing on: When someone sits down at a piano for the first time and the song comes out ugly, the instrument isn’t to blame—the untrained hands are. And the same is true of AI: One clumsy first attempt says almost nothing about the tool, and almost everything about how much practice the player has put in.
I also love that comparison because playing the piano is an artistic form of delegation: You hand the instrument the job of producing sound, but it plays your vision—and, in a way, your voice. Sit down without either one, pound the keys, and the problem isn’t the piano. It’s the player—and the same is true of every AI draft that disappoints you.
So delegate first drafts to AI when it’s appropriate, but always remember that your vision and voice are what ultimately turn random sounds into beautiful songs.
Do
Once your first two groups are humming, it’s tempting to keep going—to hand AI this task and that task and the other task, and more and more and more. But the list ends exactly where it should: at things that only you can do.
The do group is the relational, self-improvement, or high-stakes work that humans hold on to—because AI shouldn’t do it, because AI can’t do it, or because doing it is the whole point.
The test for this group is different from the first two, and the difference matters—because this time, only one answer to any of the questions below needs to be “yes.”
- Is the task relational (does it require your presence, your eye contact, or your in-the-room judgment)?
- Is it high-stakes (too high to hand off, even with review)?
- Or is it for self-improvement (where the doing itself is the point because the process produces the benefit)?
Again, one “yes” here is enough—because any task that involves relationships, high stakes, or self-improvement has to stay in human hands.
That said, let’s think about what your follow-up engine and its carefully crafted emails are actually protecting: your time. Every hour they save you is an hour you win back for the work only you can do. With that time, you can be fully present when a young family walks through your showroom door. You can sit down for the consultation at that family’s kitchen table, where you’re not so much selling a product as solving a problem. And you can hand-write the thank-you card that shows up in the mailbox a week after the install—the one no algorithm on earth can fake. Moments like those can’t be automated, and they shouldn’t be delegated, because they’re the reason the first two groups exist at all: You hand off the mundane so you can be unhurried with the meaningful.
You’ll also notice that this section doesn’t end with a prompt—because there’s no such thing as a prompt for being present.
There is, however, a warning worth taking to heart. Back in 2008—long before anyone had heard of ChatGPT—Nicholas Carr asked a famous question in the Atlantic: “Is Google Making Us Stupid?” The internet, Carr worried, was turning us into what playwright Richard Foreman called “pancake people”—intellects spread wide and thin. Carr concluded that article with a sentence that rings truer every year: “[A]s we come to rely on computers to mediate our understanding of the world,” he wrote, “it is our own intelligence that flattens into artificial intelligence.”
Read that sentence again, and notice which direction the flattening runs: The machines don’t rise to our level; we sink to theirs—but only when we hand them the work that was keeping us sharp. That’s what’s truly at stake in the do group. Give away the wrong tasks—relationship building, deep thinking, and self-improvement—and you won’t just lose a few sales. You’ll slowly flatten into a pancake person (or worse, maybe a crepe).
So automate the repetitive, and delegate the drafts—but hold on to the do group with everything you’ve got.
Write That Down
At this point, let’s return one last time to that lecture hall in The Office—only now, don’t watch Michael. Watch the students. Again, while he lectures foolishly, they type ferociously, carefully recording every word of his errant lesson.
If you’ve spent any time listening to the loudest voices in conversations about AI, you’ve seen that joke play out in real life: One expert says AI is a worthless fad, another swears it can run your whole company without you—and thousands of smart, busy business owners type those wrong lessons straight into their laptops.
But here’s the hard truth that both of those voices miss: AI definitely won’t save the world, and it probably won’t destroy it (though I suppose the latter’s possible). Yet when it’s used with discernment, it can free up humans to focus on what matters most.
So what does that kind of discernment look like inside a hearth shop? It looks like one sale, handled by the whole system: Your follow-up engine nudges a warm lead before it goes cold. The email that brings that customer back to your showroom sounds exactly like you because AI drafted it and you approved it. And the sale closes at a kitchen table, where you’re fully present, solving a family’s problem.
Automate.
Delegate.
Do.
That’s the power of this framework—it lets AI handle the mundane so we can pursue the meaningful.
With all that in mind, here’s my challenge for you—and fittingly, it starts with a piece of paper. Grab a pen (not a keyboard), and write down three things: one task you’ll automate this month, one task you’ll delegate this week, and one human moment you’ll be fully present for with the hours you get back.
The students in Michael’s classroom typed the wrong lesson into their laptops. But your takeaways will be different: three lines on real paper in your own hand.
Write that down.
Want to get free content like this every month?
Matt Bradley
Matt Bradley is the partnership manager at WhyFire and the editor of The Fire Time Magazine.
