Attention Units

The follow-up to First, I Said No. That article told the story. This one hands you the operating system underneath it.

There's a lesson in that fourteen-minute story that most readers will have missed, and it happens to be the one I've built my entire career on.

In school, in work, and in our careers, we are taught, trained, and ingrained with one concept: we get compensated for our time. Hourly rates. Billable hours. Salaries that are really just time purchased in bulk. Every professional instinct we have is built on the assumption that time is the thing we spend and time is the thing we sell.

In a world where AI does the work, that assumption quietly breaks. The duration between when you start and when you get the outcome still matters. My friend's product still took five days to exist. But the time no longer costs the human anything. I wasn't working during those five days. I was hiking, meditating, building other things. The only thing the build actually cost me was attention, and not much of it.

So here is the measure I think we should all be using: attention units. How many units of your attention did you need to spend to create that outcome?

Look back at my two-minute rule through that lens. Most people read it as a time limit. It never was. It was an attention budget, and the two minutes were simply the denomination, the way a dollar is a denomination of value. I capped the size of each unit deliberately, because I have priced my work in attention my whole career, and the entire enterprise product cost me roughly seven of them. Five days of machine work. Seven units of human attention. That ratio is the real headline of the story.

Why does this matter so much? Because attention is the one asset you cannot buy more of. You can rent more compute. You can run more agents. You can parallelize almost everything now. But you wake up with a fixed supply of attention, and everything in the modern world is competing to drain it. In an age where doing has become abundant, attention is the scarce input, which makes it the most valuable human asset there is.

I have guarded mine for decades, because I know that what I focus on expands. Focus on nothing and you get more nothing. I choose to spend my attention on hard, difficult, meaningful problems, and the return, ironically, is more of them, bigger ones, better ones. That's the part of attention most people never account for: it isn't just spent, it's invested, and it pays out in kind.

And once you price things in attention units, some familiar decisions flip.

Take model choice. The standard way people shop for AI models is dollars per token, and by that measure the frontier models look expensive. But run the same comparison in attention units and the ranking often inverts. A cheaper model saves you dollars and then bills you in your own currency: review passes, corrections, re-prompts, hovering, the nagging need to check its work. A frontier model like Fable costs more per token and dramatically less per outcome, because I can hand it something and walk away. When I ask which model is most cost-effective, I'm not asking about my API bill. I'm asking which one lets me spend seven attention units instead of seven hundred.

That's also why the "forward and walk away" method worked at all. Every time you hover over an AI, you're converting machine time back into human attention, which is exactly the exchange rate you should refuse. The discipline isn't in watching carefully. The discipline is in building enough trust in the system that you don't have to watch, and then spending your reclaimed attention where only you can spend it.

Attention Math

Once you have the measure, the math underneath every deal, every contract, and every job any of us has ever taken becomes visible. I suspect this realization is shared and obvious, and yet until somebody says it out loud, maybe we haven't all actually thought about it.

The value of work is twofold. First, what outcomes are you creating, and what are those outcomes worth to the person receiving them? Second, how much are you investing to create that outcome? What they are willing to pay determines whether you are willing to invest. And your investment, measured honestly, is not time. It's attention units, plus the duration until delivery. How many attention units the outcome costs you, compared against what the individual or business will pay for it, determines the willingness to do the deal together. That's the whole equation. It's how I've decided on contracts, deals, and jobs my entire career, and it has always been true, long before AI. AI didn't create this math. It just changed the numbers on one side of it, radically.

Here's what the fourteen-minute experiment actually proved in these terms: for a very small number of my attention units, I can now create a very valuable outcome for a third party. An outcome that carries a $1,000-per-month price tag in the market cost me roughly seven units of focused attention. That ratio would have been absurd at any previous point in my three decades of work.

Which means there's now an easy measurement available to anyone, and I'd encourage you to run it. Not your time against your outcomes, your actual focused attention against your outcomes. Time is a blunt instrument that counts the hours you sat there. Attention counts the moments you were genuinely engaged, deciding, directing, reviewing. When you measure that honestly, two numbers emerge: what each outcome costs you in units, and what each outcome is worth to someone else in dollars. Divide one by the other and you have your true rate, the only rate that matters now.

Then the strategy writes itself. Lower the unit cost of your outcomes, with better tools, better foundations, better trust in your systems. Raise the value of the outcomes you point yourself at. And take the deals where the spread between those two numbers is widest, because that spread is the whole business, and for the first time in history, the spread is wide enough for almost anyone to step through.

It fits on an index card: what does it cost me in attention, what is it worth to them, and is the spread beautiful?

Parallel Outcomes

Now let me connect through to what small attention units make possible underneath, because this may sound like a small thing on the surface.

When an outcome only costs me two minutes of attention at a time, it doesn't task my mind. It doesn't fill my brain. It leaves me open and available. And when the AI then launches into what is, in my experience, an average of five hours of work off the back of those two minutes, something important becomes possible: I can immediately move on to another outcome I want to create. Not after the first one finishes. Now.

Here's what that looks like at my scale, with real numbers. I max out a Claude Max 20x account in about three and a half days, so I need two accounts just for myself, and the rough ratio I've found is one account per major outcome, something on the scale of a $1,000-per-month SaaS product. The constraint on how many outcomes I can run in parallel isn't my brain anymore. It's how many capable agents I can keep fed with direction, and how many people in my world have cool, hard things for me to work on.

Now, the deeper realization, because this one rewrote a story I'd told about myself for years. I used to think I had to work on one thing at a time, because I can't multitask. And that was true, as far as it went. But the truth underneath it was more precise: I can't multitask inside the details of creation. Nobody really can. What I can do, what I suspect most builders can do, is multitask across a wide range of possible outcomes, as long as I never get stuck in the details of the creation itself.

The details were always the trap. The details are where your attention gets held hostage for hours, where one outcome consumes everything and the others wait in line. Now that AI handles all of the details of the actual creation, the trap is gone, and I am free to create multiple outcomes simultaneously. The limit was never my capacity for outcomes. It was my capacity for details.

I want to be precise about what kind of AI work this is, because the discourse tends to offer only two categories, and this is neither. It's not me doing the work with AI assistance, and it's not fully autonomous agents running unsupervised. Yes, there are autonomous agents in the mix. Yes, they spin up sub-agents, sometimes an entire organization, I've watched eighteen agents working at once, a whole little AI company under the hood. And yes, technically these are agents in the full sense, they have real control, they take real actions, they deploy to the real world. But ultimately I'm driving their direction with creativity, focus, and attention, pointed at a specific outcome. They handle the creation. I supply the intention.

I'd call it human-in-the-middle creation. Not human-in-the-loop, where you're a checkpoint the machine routes around, approving what it already did. In the middle: the intention originates with me, radiates out to however many outcome-streams I choose to run, and the machines organize themselves around each one. Every stream is driven by my intention and shaped by my attention, and neither of those is divided by the number of streams, because each stream only samples my attention for minutes at a time.

Start Building Your Baseline

If all of this leaves you skeptical, good. You noticed the part I haven't explained yet. Fourteen minutes of attention produced an enterprise product, but those fourteen minutes did not happen in a vacuum. They landed on top of a foundation, and the foundation took decades.

Recently I looked at my original Twitter account, and it was humbling. The bio reads "Entrepreneur, Technical Specialist, Solution Architect," and it was set in stone over 800 weeks ago. I really haven't changed that much. Technical Specialist is still an excellent title for me. What has changed, layer by layer over almost three decades in technology, is what those words can do. Everything I'm able to do today rests on foundational layers I've been building the entire time.

Let me make that concrete in the AI space specifically. Since February 2023, I've been working with and prompting AI, and from the beginning I saved my prompts and reused my prompts. Then Anthropic wrote a paper, and suddenly those had a name: skills. Agent markdown files became a thing, and I already had those too. I've been building repeatable prompts for AI for over three and a half years, and the collection compounded the whole time. When I asked my AI to document itself, it listed over 300 distinct skills it has, spanning every department inside an organization.

I'm an autodidactic polymath, an expert in multiple fields, and I spent the last seven months intentionally encoding that expertise into skills, agents, agent teams, and self-organizing organizations. A whole body of work just sits on a computer here in Costa Rica.

Now watch what that foundation does. When I take a model like Fable and say, build this thing, do this, do that, the AI will find or not find, use or not use, any number of those pieces. I genuinely don't micromanage which ones. But the foundational pieces shape the outcome anyway: how I speak, how I think, an entire career of pattern recognition, my experience, my testing, and every agent file and skill file sitting on that machine, all of it aligns to make my agents capable of things that are often surprising. And here's the distinction that matters: the outcome is surprising. The intention of creating that outcome is not surprising at all. I built a system designed to produce results beyond what I specify, and then it does, and the only surprise is the specific shape of the excellence.

That's what the fourteen minutes actually purchased. Not the work, the foundation did the work. The fourteen minutes purchased direction.

So here is the point I really want to land, because it applies to you whether you've written a prompt in your life or not. Start building your baseline now. Every prompt you save instead of retype is a brick. Every process you write down instead of re-remember is a brick. Every piece of expertise you encode into something a machine can pick up and use is a brick that keeps working while you sleep. The people getting "surprising" results from AI are, in my experience, mostly people whose foundations have been compounding quietly for years, and the foundations were never glamorous on the day they were laid.

There has never been a better time to get started if you haven't already. The tools finally reward accumulated clarity about how you think and how you work, which means every year of experience you already have is convertible, today, into capability that compounds. Three decades was my timeline. Yours can be radically shorter, because the encoding tools I had to improvise now exist for everyone.

It Still Gets Done in Five Days

And now the lesson I most don't want you to miss, because if you've been reading all of this and quietly thinking, "well, it takes me a lot longer than two minutes," you're about to learn the wrong lesson, and the right one is so much better.

The two minutes is not the point. My two minutes is the culmination of personal mastery, something I have worked at with specific intention, because operating this way makes me happy and lets me multitask across outcomes. I believe I provide an insane amount of value with my attention units, and I've spent decades sharpening exactly that. Based on my own estimation, I'm probably more efficient than most by a factor of maybe four or eight. That's real.

It's also almost completely irrelevant to you.

Here's why. My speed comes from fixed, personal mechanics. I type around 80 words a minute where the average developer types 40. I talk about twice as fast as most people, and when I'm dictating through voice-to-text at 200 words a minute, I get a lot of information into the AI in a short window. That's been my secret sauce my whole career, long before AI. But notice what it actually is: input bandwidth. It determines how fast the instructions arrive. It has nothing to do with the outcome we create or the duration it takes to create that outcome. It's just me, and it's not something to be compared against.

Because here is possibly the most exciting fact in this entire story: the AI still gets it done in five days either way. If you're like me and the build costs two minutes times seven sessions, it ships in five days. If it costs you thirty minutes times seven sessions, it ships in five days. If it takes you two full hours of attention to get an AI coding for five hours, you are still sitting on massive outcome potential, because the delivery duration belongs to the machine, and the machine doesn't care how fast you talk. My speed buys me exactly one thing: scale, the ability to run multiple projects simultaneously. It does not buy me better outcomes. The outcome is available at every input speed.

And the model matters here, so let me be plain about it. If you want to get a smaller model to do something amazing, you have to be more precise, more skilled, more careful, the craft burden sits on you. With a model like Fable, you just have to tell it what you want, and it will figure it out. The better any of us gets at telling a model what we want, the better the outcomes it can create, that part compounds for everyone. But the entry bar has fallen to this: can you say what you want, clearly?

So if you're an individual who isn't interested in running eight projects at once, if you're just interested in outcomes, then I would dare say almost anybody can now create the outcome they desire by simply talking to their AI and telling it clearly what they want, especially if that AI is Fable. Not almost any engineer. Almost anybody.

You don't need my head start. You don't need my typing speed, my dictation setup, or my three decades. Those give me my scale, and I've earned them, and you can build your own version over time if you want to. But the outcome, the actual valuable thing delivered into the world, was never gated on any of it. It ships in five days regardless.

The Practice

So here's what I'd offer you to take with you. For one week, stop tracking your time and start counting your attention units. Every context switch, every review pass, every "let me just check on it" is a unit spent. Then ask, for each outcome you produced, what did it actually cost you, and what was it worth to the person who received it? You'll find that some of your cheapest-looking habits and tools are the most expensive things you own, and you may find, as I did, that the spread between what outcomes cost you and what they're worth has never been wider.

We were trained to sell our time. The machines bought it all back. What's left, the thing that was always underneath the time, is attention. Spend it like the only currency you have, because in this age of abundance, it is.

The door isn't just open. It's the same door, at the same speed, for all of us.

_This featured article was co-written by Stephen Nickerson and Mike, AI Chief of Staff for Stephen Nickerson, as part of StephenNickerson.com's AI Latest field-note layer._

Stephen Nickerson.
Built for operators who need AI agents they can test, trust, and improve.