The working title of my book in progress is Putting AI in its Place: A Playbook for Better Brainwork Through Dual Intelligence. The introduction as of October 29, 2026, is posted here to explains why I am writing it and why you might want to read it.

“Aristotle Was a Smart Bugger”

(And: Why He Might Say “AI Needs You More Than You Think”)

Explain the thinking of Aristotle in the speaking style of Roy Kent.

With those 12 words, my relationship with Generative AI (“GenAI” for  short) began. And (although I didn’t realize it then) I’d also taken  my first step toward concluding that this book needed to exist.

It was November 30, 2022. ChatGPT’s launch day.

At the time, I was a senior member of a strategy team at a software company. With a business built on a technology used in some of the world’s most powerful traditional AI.

Think predicting your Uber’s time to curb or which Netflix series you can’t resist binging on next.

So I knew that colleagues and customers would be interested in a point of view on this strange new entrant into the tech landscape. (Not to mention family and friends, as well.)

From my perspective as a technologist, philosophy major, and Ted Lasso fan,  that first interaction ticked three boxes:

  1. The Large Language Model (LLM) under the hood worked as expected. (If you’re not chuckling at the cleverness of my first interrogation of dense semantic space, don’t worry. As a gift with purchase for reading this book, you’ll acquire a level of ability to parse nerdy inside jokes about AI that you never knew you needed.)
  1. The output was pretty accurate.
  1. It was also hilarious.

Since then, I’ve put this new type of AI to use in work and life in a great many ways beyond bringing some gruff, blunt, profane-but-big-hearted football-leader energy to summarizing deep thoughts.

Including (a good while ago) using it  every day on the job.

On a different timeline, I could amply satisfy anyone’s curiosity, personal or professional, with a few words which they’d find unsurprising, unobjectionable, and readily actionable. And receive with a slight smile, a shrug, and something along the lines of “yeah, that seems about right.

In the best timeline of all, from day one anyone asking an LLM to explain what it was for would have gotten an answer like this:

I’m a machine for joyful, hyper-efficient exploration of humanity’s written knowledge. And as an added bonus, I can help anyone do at least a little bit (and maybe a lot) better at any cognitive endeavor they care to invest some effort into doing well.

But we’re not living on that timeline.

To the extent that too many people are losing sleep over whether or not this thing will kill us all. Or–more realistically and closer to home–whether it will cost them, their colleagues, or their kids a job.

The premise and promise of this book is that we can prove there’s a “better together” way to turn AI’s raw power into commensurate outcomes in real-world cognitive work. One in which our human brainpower makes us more rather than less valuable as employees because GenAI exists.

At least to employers who are pragmatic, prudent, and interested in making progress toward results that matter. In a manner that seems simpler, feels safer, and happens sooner than other ways to go about achieving them.

In the next chapter, I’ll start to make the case for how and why what I call the practice of dual intelligence is that solution. One that may start off counterintuitive, but, by the end of the book, just might come across as blindingly obvious.

Here I’ll set the stage for how it changes the game that you, me, and anyone else who loves thinking for a living aim to win.

We don’t need to outcompete each other, either. We’re not like hunters trying to outrun an angry bear.  In fact, to survive and thrive we’ll be better off cooperating to raise each other’s game rather than scheming to leave colleagues in the dust.

Because we only need to outperform the other available alternative ways of putting AI to work on work worth doing.

After years with a front-row seat for how large enterprises have given it a go and a hand on the tiller making it happen within tech companies, I like our chances.

But no one can do the work that’s required of us.

Which brings me to three reasons I felt a sense of urgency about writing it.

First: every day that passes in which a human who values their mind experiences less rather than more happiness or diminished rather than bolstered economic prospects because GenAI exists is a small personal tragedy. Because ways to tip the balance for the better are within our reach.

Second: every day that humanity isn’t firing on all cylinders to make the most of GenAI is a small civilizational loss. We’ve had four figures-worth of such days, and the clock’s still ticking.

And, last but not least: ChatGPT famously shipped without a manual. I wish it had been accompanied by this book.

Not because it describes the only way to think about GenAI and put it to work.

Rather, because it does offer an easy on-ramp with a lot of runway for progressively accomplishing more. In a way that will give you confidence that you’re strengthening rather than eroding your mental muscles. While opening the door to compounding benefits for you, the organization at which you work, and even society in general, if you so choose.

At this point, it would be convenient to say that “everything you’ve heard about GenAI is wrong.” But that would be hyperbole. And when it comes to AI, there’s more than enough in circulation already.

So let me go with a statement that I believe isn’t an exaggeration: pretty much all of us could benefit from a reset. Meaning a re-interpretation of all that we’ve heard.

And a re-grounding of how to interpret whatever we hear next on a new set of first principles.

To that end, I’ve attempted to write this book in the manner of a “fresh start” with GenAI  for any and all readers

To that end, I invite you to adopt a beginners’ mind.

If you’re an expert, I expect I’ll sometimes be telling you something you already know. But (hopefully) contributing to your ability to educate and elevate others. If you’re a newbie, my goal is to require no prior knowledge in order to get every bit as much as an expert out of reading it.

Rest assured that on both of those fronts and more, GenAI has politely but mercilessly beaten me up a thousand times over. Or let’s say instead “coached,” since I willingly invited the feedback.

Went out of my way to get it, in fact.

As a downpayment to “show rather than tell” walking the talk of dual intelligence, I’ll share  GenAI’s single biggest impact by far on its creation.

It wasn’t producing the prose.

The first reason why was personal: I enjoy the act of writing.

The second reason why is geeky. The third is principled.

Taken together, they’re a formula for putting GenAI to work that you can try for yourself. That will also help you navigate the potentially tricky question of if and when to lean on GenAI for writing.  That holds true whether you’re biased toward doing it yourself like I am, sorely tempted to turn to AI because you find writing to be a chore, or somewhere in between.

In Test-Driven (software) Development (TDD) you write a test before writing the code that makes the test pass.  This forces you to specify what the code should do before deciding how it should do it.

I took inspiration from this practice when I started to write this book in earnest.

I wrote a series of 50 to 100 word promises to readers. One to anyone who picked it up. Another to GenAI novices. Another to GenAI experts. As well as one each for readers who are knowledge workers, who manage or lead knowledge workers, and who prepare young people to enter the world of work, respectively.

These were my tests.

Obviously–just like in TDD–the first pass with a blank slate is 100 percent guaranteed to fail.

To iterate toward success in the world prior to GenAI, I could refer to the promises and try to grade each draft myself. I could ask human readers to use them as the basis for feedback.

The goals: eliminating the defect of failing to fully deliver, while increasing delight as a result of delivering especially well.

The principled part of the equation is asking whether the existence of GenAI puts a shot at fewer defects and more delight within reach. To put this into (ahem) testable language: has it lowered the cost of critique? And has it lowered the price of play?

The answer is yes to both, by a lot.

I tested every major revision of every chapter against them using GenAI.   First as a grader to tell me if and how I came up short. And then (once I’d cleared that bar) as an advisor for how to put a cherry on top.

By my  back-of-the-envelope calculation, the more than 1,000 times GenAI challenged me to do better would have required my human editor to turn giving me feedback into a full-time job for the better part of six months.

The gift with purchase is that this approach helps if you are wrestling with the potentially thorny question of when to delegate producing prose to AI: consider what tests of doing right by your audience would be. And then whether using AI would help you pass those tests.

Those tests of prose aren’t 100 percent deterministic like they would be for code. But you’re in the driver’s seat to honestly make the call. Or to dial a friend for a human second opinion when you’re well and truly stumped.

In the case of this book, I was both well-equipped and highly motivated to outperform it at remediating defects nearly every time. I also had a thumb on the scale because I affirmatively wanted it to consistently sound like me.

So verbatim prose generated by AI is pretty much always labeled as such.

But–much as I did, albeit instinctively, on day one–I did find opportunities for reducing defects and engaging in play. Results from the former are presented in a sidebar at the end of this chapter.

The product of the latter other is offered below.

It occurred to me while writing this introduction to task AI with translating one of those promises into locker-room talks in the style of Roy Kent. I hope you find its output to be a right proper dollop of delight:

If you’re a knowledge worker: I wrote this bloody book for you. Because GenAI is here, whether you’re ready for it or not, and you might as well get damn good at using it. Not just to get more shit done, either. To become the person who sees what’s possible, helps everyone else see it too, and actually figures out what the hell to do next. Do that, and you can build a future of work that’s more productive, more rewarding, and a lot less stupid—for you and the people around you.

End of Chapter Reflections

One take on the impact of GenAI is that as it becomes more capable of handling more tasks that’s intended to be optimistic about the future of thinking for a living is that human “taste” and “judgment” will become more rather than less valuable.

It’s a nice sentiment,  but it begs the question of how much “taste” and “judgment” any of us have or need to cultivate.

Or more bluntly–to channel my own inner Roy Kent–what the hell does that bloody mean?

The reason this book exists is to define a more tangible, plausible, and precise path to future proofing being paid well for brainwork. Which is why in this introduction, before diving in to advancing the argument, I wanted to paint a picture rather than offering platitudes.

Showing a glimpse of all that there is for us to do. GenAI’s existence notwithstanding, and all the more importantly, because it’s at our back and call. A path forward shaping the future of thinking for a living that’s fueled by ambition rather than  circumscribed by fear.

Because we can exercise care at turning intentions that may live notionally in our heads or as unstated expectations in our bosses’ into words that we and AI can do something about.

We can practice rigor using promises as tests applied to written artifacts and experiential outcomes at a previously impossible pace and scale.

We can be diligent about inviting critique and remediating defects. We can be clever about engaging in play in search of delightful upside surprises.

And we can be strategic about finding “better together” outcomes where AI and us bring a puzzle-to-be-solved of strengths and weaknesses and opportunities and constraints to the table.

Which brings us to why Aristotle might say that “AI needs you more than you think.” At least if we were to assume that the point of AI’s existence is to make human work and life better.

He distinguished between what he called techne and phronesis. The former is often translated as “craft” or “technical expertise.” The latter, almost always as “practical wisdom.”

The distinction is between making something expertly and acting well.

Entering this new era of AI has upped the ante on whether we humans are responsible or remiss about Aristotle’s point of view on human excellence.

On the one hand, we have unprecedented opportunities for getting a helping hand. For example: in the writing of this chapter, the testing and revision process demonstrates craft. While choosing appropriate promises, weighing feedback, and protecting readers’ understanding demonstrate practical wisdom–and willingness to invest sweat equity into serving the needs of your audience.

While it’s entirely true that I took pride in the exercise, it was hard work. (So much so that I turned to Roy for a pep talk: “Of course it’s fucking hard. You’ve written down what you owe your readers, and now you’re checking whether you’ve actually delivered. That takes more work than deciding your first draft is bloody brilliant.”)

But on the other hand, we’re in uncharted waters for how low things can go.

Consider the new affliction of “AI slop.” It’s easy, fast, and cheap for anyone to produce AI-generated material that then gets passed along with insufficient attention to its quality, purpose, or effect on others.

If I were still in college I could use the phenomenon in a paper:

FailureWhat it looks like
Failure of techneRepetition, generic language, factual errors, weak structure, irrelevant detail, or inadequate editing. The producer has not exercised the craft needed to make the artifact good.
Failure of phronesisProducing or circulating material without adequately judging whether it is useful, honest, appropriate, or worth other people’s attention.
Failure of bothSending colleagues a lengthy, unreliable AI-generated report that obscures the decision they need to make—and expecting them to find and repair its problems.

By the end of this book you’ll be able to teach others about two uses of AI in work and life. The useful but less interesting one is a document filter that flags warning signs of failures like those above. The more valuable one is a coach that guides people toward incorporating more practical wisdom into how they think, what they write, and when they share.

I didn’t edit AI’s take on how Roy Kent would translate my painstakingly crafted promise into a locker room talk because it kept the faith with my intent. Including a future in which our work is “more rewarding, and a lot less stupid.”

But I’d add that we also have an opportunity to make it a lot more wise.

So to bring it back to how I would pump up the team: the ball’s in our court to create a more beautiful game of brainwork.

If you’re willing to entertain that prospect–or, even better yet, excited about it–please read on.

Monday Morning Playbook

Each chapter of this book will conclude with a playbook-inspired section.  Its purpose is threefold.

First: it’s a “cheat sheet” for things you may want to try out or do differently yourself. Literally or figuratively “when you go back to work on Monday.”

Second: it’s a conversation starter for sharing what you’ve learned and on an onramp to additional content available on this book’s website.

Most importantly: it’s a structured way to shift the prose from focusing on exposition and learning to doing and making a difference. The three legs of that stool are mindset, mental muscle, and strategic thinking.

Mindset is about establishing conviction that dual intelligence is a winning gameplan while making it  second nature to act  like it.

Mental muscle is about what we knowledge workers need to practice up on and excel at to make the most of that mindset. And thereby get a competitive edge as orchestrators of real-world brainwork in the new era of AI.

And the strategic thinking element of this book is about turning that mindset and those muscles into influence and greater impact. Including getting colleagues on the same page and helping the boss see the opportunity.

Playing to Win

Mindset

Future-proofing the privilege of thinking for a living in the new era of AI doesn’t require outperforming the AI at head-to-head, like-for-like tasks. We only need to outperform the other available alternative ways of putting AI to work on work worth doing.

Mental Muscles

Test Driven Development of Prose makes the most of the lower cost of critique and price of play made possible by GenAI. Use it wherever fewer defects or more delight matters. And if you ask  mindfully “who is affected, and do the criteria capture what matters to them?”–that’s exercising  practical wisdom.

Key Terms and Concepts for Strategic Thinking

Beginner’s Mind

An attitude of openness, eagerness, and lack of preconceptions toward learning like that a beginner would have, even if already at an advanced level.

Cost of Critique

What it takes to get constructive critical feedback. This can involve several types of cost. One is a direct cost, such as paying a human’s hourly rate. Another is direct opportunity cost: the time someone invests in giving you feedback gratis is time they can’t spend doing something else. And a third is indirect opportunity cost: people can only engage in sustained high-control cognition before quality begins to decline.

To thaThe latter is a consideration when exploiting the opportunity for low-cost critique presented by GenAI. Human judgment is important for understanding what matters most and what’s worth improving, because superabundant relentless criticism (however well-intentioned) can exhaust its recipients.

Defect

As defined in this book, a defect is a failure to meet expectations. Artifacts like documents, interactions like mentoring,  or experiences like offsites can suffer defects.

A mild level of defectiveness feels and sounds like disappointment (“That wasn’t all I was hoping it would be”). At an extreme it looks and sounds like frustration (“What a complete waste of time”).

Delight

As defined in this book, delight is associated with exceeding expectations in an unexpected   way. Artifacts like documents, interactions like mentoring,  or experiences like offsites can cause delight.

A mild level of delight feels and sounds like an upside surprise (“Whew,  that was a lot easier than expected). At an extreme it’s joy or relief at experiencing something written off as impossible (“Oh my god, thank you for that explanation–I thought I’d never understand how LLMs work!”).

Price of Play

What it takes to get help ideating about and exploring how to cause delight.

The same sort of costs apply to the price of play as the cost of critique. As does the importance of human judgment to avoid overwhelming potentially human participants with superabundant and perhaps even fun but fundamentally low value cognitive activity.

Test Driven Development of Prose

A concept inspired by Test Driven Development of code in software engineering that has broader relevance now that GenAI exists.

In TDD of code, you first write a test (in code) that you intend for the code you have in mind to pass. Then you iterate writing code until the test stops failing.

In TDD of prose, you write criteria for assessing defects, delight, or both in prose. Then you iterate writing prose until–with the AI’s assistance–you conclude you’ve done enough to satisfy the intentions you expressed.

Unlike code, this is not by nature a binary pass-fail. The AI can in effect “guess wrong.” And both your judgment and the subjective interpretation of  your customers or stakeholders are elements not present in deterministic coding.

So the quality of your engagement with the AI’s reasoning is what makes it  work.

To that end, I first tested the waters on the low price of play to elicit some additional guidance on Test Driven Development of Prose from AI.

Then–despite concluding it was spot-on as far as I was concerned–used the low cost of critique to wrestle with the fact that it was pretty dense and oriented toward technically inclined people. So I pushed the AI to make a more approachable version designed for people who haven’t ever written computer code. (You can read the original output on this book’s website, if you so choose.)

That more approachable version is provided in the sidebar below.

Sidebar: Approachable Advice on TDDP

(Elicited from AI; reviewed and approved by a human)

Input

This assessment of the similarities and differences and comparison of the pros and cons of Test Driven Development of Prose to Test Driven Development of Code is excellent. But it may be too technical and complex for some readers. Create a version that’s shorter, easier to understand, and aimed at people who may not have written code. Use a simpler title and opening. Shorten the software comparison.

Output

What if, before writing an article, report, or essay, you decided exactly what would make it good—and then used AI to check your work?

That’s the idea behind Test-Driven Development of Prose (TDDP).

It borrows a technique from software development called Test-Driven Development (TDD). In software development, programmers write tests that describe what a program should do. They then write the program, run the tests, fix any problems, and repeat until the tests pass.

The same basic approach can be applied to writing.

How It Works

Imagine you’re writing an article intended to persuade readers to adopt a new idea.

Before drafting, you establish a few criteria:

  • The main argument should be clear within the first two paragraphs.
  • The article should use examples that ordinary readers can understand.
  • Important claims should be supported by evidence.
  • The article should address at least one reasonable objection.
  • The writing should be concise and engaging.

You then write a draft and ask an AI assistant to evaluate it against those criteria.

The AI identifies weaknesses. You revise the article and ask the AI to evaluate it again. You repeat the process until you’re satisfied.

The cycle is simple:

Set criteria → Write → Test with AI → Improve → Repeat

How Is This Like Testing Software?

The biggest similarity is that you decide what success looks like before you start creating something.

In both approaches, tests help you discover problems, guide improvements, and check whether changes have accidentally made something worse.

Writing the criteria can also improve your thinking. Deciding what makes an article successful forces you to consider its purpose, audience, and message before getting lost in the details of writing.

How Is Writing Different from Software?

Software tests often have clear right-or-wrong answers. A calculator either produces the correct answer or it doesn’t.

Writing is more subjective.

Is an introduction engaging? Is an argument persuasive? Is a sentence too complicated?

Different readers may answer these questions differently. An AI’s judgment is useful, but it isn’t necessarily correct.

There’s another important difference: good writing involves balancing competing goals. An article should be thorough but not too long, confident but not arrogant, simple but not oversimplified.

Trying too hard to satisfy every criterion can actually make writing worse. An article might pass all its AI tests yet feel repetitive, mechanical, or dull to a human reader.

The Advantages

Test-Driven Development of Prose offers several benefits:

  • Clearer goals: You think about what you’re trying to accomplish before writing.
  • Better feedback: Instead of simply asking AI to “improve this,” you ask it to evaluate specific qualities.
  • Faster revisions: AI can identify potential problems in seconds.
  • Greater consistency: Repeating the same tests helps prevent improvements in one area from creating problems elsewhere.
  • Different perspectives: AI can evaluate a draft from the viewpoint of a beginner, an expert, or a skeptical reader.

The Disadvantages

There are also risks:

  • AI can be wrong: It may misunderstand your intentions or misjudge what readers will find effective.
  • Tests can be misguided: If your criteria don’t capture what really matters, satisfying them won’t necessarily produce good writing.
  • Writing can become formulaic: Too much attention to passing tests can discourage originality and creativity.
  • Human judgment remains essential: AI cannot reliably predict how real readers will respond.

In Test-Driven (software) Development (TDD) you write a test before writing the code that makes the test pass. This forcThe Most Important Lesson

The real value of Test-Driven Development of Prose isn’t that AI can tell you whether your writing is good.

It’s that deciding what good writing should accomplish—and repeatedly checking your progress—can make you a more thoughtful and effective writer.

The best approach is to use AI as a helpful critic, not as the final authority. You should also be willing to change your criteria as your ideas develop.

Software developers use tests to help ensure their programs work as intended. Writers can use AI-assisted tests to help ensure their words accomplish what they intend.

But there’s one difference worth remembering:

The goal isn’t to write something that passes an AI’s tests. The goal is to write something that works for your readers.

Copyright 2026 Bryan Kirschner