Limitless

In Limitless, Bradley Cooper plays a struggling writer who discovers NZT-48 — a pill that unlocks his full cognitive capacity. He finishes his novel in four days. Masters the stock market in a week. Learns Italian in an afternoon. The pill doesn't give him new knowledge. It gives him access to everything he already had — every book he'd read, every conversation he'd overheard, every connection his brain had already made but couldn't surface.

NZT didn't make him smarter. It made him fully himself.

I've been thinking about this movie a lot lately, because we're now living in a world where we can actually measure the before and after. Not with brain scans or IQ tests, but with something the AI industry gave us a perfectly good unit for: tokens.

What if we audited human work output the same way we price AI output? What does a knowledge worker actually produce in a month? What does it cost? And what happens when you hand them the NZT-48 of a $20/month AI subscription?

The answers are funnier, sadder, and more economically absurd than I expected.

Section I: Your Daily Token Receipt

Let's start with a provocation: if we priced your work output like AI output, what would the invoice look like?

A "token" in AI is roughly 0.75 words. Every email you write, every Slack message, every sentence you speak in a meeting, every line of a document — it all converts to tokens. So I audited a typical day for a Software Engineer and added up every word that leaves your brain through your fingers or mouth.

That receipt is built from observable activities: 10 distinct output events across an 8-hour day, plus the after-hours bleed of couch emails, Sunday night prep, and commute voice notes that nobody pays you for. The daily tokens, multiplied by 22 working days, give us your monthly output: 0 tokens.

And here's the punchline nobody asked for:

Your Context Window is Embarrassingly Small

It gets worse. Miller's Law (1956) tells us humans hold 4-7 items in working memory at once. In token terms, that's roughly a 4K-8K context window. With notes, tabs, and tools open, maybe you stretch to 16K-32K effective context.

You're running a 4K context window on a $113K/year salary. Claude runs 200K for $25 per million output tokens. The reason you're still employed is that your 4K tokens carry context, relationships, organizational knowledge, and judgment that no context window can replicate. Your tokens are expensive because they're right.

Not All Humans Produce Equal Tokens

The gap between top and bottom performers isn't just volume — it's signal density. Top performers don't send more emails. They send fewer, better ones. Their meetings have less filler. Their documents resolve issues instead of creating them.

Notice the top 10% produce ~41% more tokens, but the real gap is larger. Their meeting filler ratio is ~30% vs. ~60% for the bottom 10%. Their emails resolve issues in 1 thread instead of 5 back-and-forth chains. Token efficiency, not volume, separates tiers.

Section II: The Price of Thinking

Here's where it gets economically absurd. Let's put your output tokens on the same pricing menu as AI models — because in 2026, that comparison is no longer theoretical.

Read that table slowly. The API cost to generate your monthly output — at the best model on earth — is less than a fancy coffee. The subscription cost is $20 flat. Your salary is $9,377.

But don't make the mistake of thinking this means humans are overpaid. The table reveals the opposite: you're not paid for token generation. You're paid for token selection. Knowing which email to write, which analysis matters, which meeting to cancel, which requirement to cut. An AI can generate 10 million tokens a day. It cannot decide which 10,000 of them actually matter.

Section III: Who Takes the Pill

In Limitless, NZT-48 was available to anyone who could find a dealer. The limiting factor wasn't access — it was willingness. Some people took the pill and built empires. Others had the pill on the table and never picked it up.

We're watching the same distribution play out right now with AI adoption.

The Psychology of Picking Up the Pill

High adopters aren't smarter. They share three psychological traits:

Why Experience Is the Real NZT

Here's the counterintuitive finding: senior people adopt AI faster and get more from it than juniors. This violates the "digital native" assumption that younger = more tech-fluent. But it makes perfect sense when you think about it through tokens.

A junior software engineer using AI generates 3x more tokens. Great. But 3x more what? If you don't know what a good test harness looks like, producing three of them faster doesn't help. You just get three bad ones quicker.

A senior software engineer with 15-20 years of experience has something AI can't replicate: a mental model of how everything fits together. They know which questions to ask, which outputs to reject, which edge cases matter, and which "best practices" are actually cargo cult rituals. When AI amplifies that judgment 3x, it's transformational. When it amplifies a junior's guessing 3x, it's just faster guessing.

The Adoption Curve: What Happens to Your Output Over Time

When a software engineer starts using AI, their token output doesn't jump overnight. It follows an S-curve that maps to learning, integration, and eventually, cognitive limits:

The plateau is the most important part of the chart. It proves AI is an amplifier, not a replacement. If AI could replace you, there'd be no ceiling — output would keep climbing forever. The fact that it flattens means you are still the limiting factor. Your judgment, your decisions, your cognitive bandwidth. AI just removed all the other bottlenecks.

Section IV: What the Extra Tokens Buy You

This is the section that matters. Not the economics. Not the psychology. This.

Because the point of AI isn't producing more. Every software engineer I've talked to says the same thing: the AI didn't make them faster at their existing work. It gave them capacity to do the work they always knew they should be doing but could never justify.

The test harness the engineer always wanted to write. The compliance review the lawyer always meant to do. The scenario model the analyst always planned to build. The patient education material the nurse knew would prevent readmissions. These weren't laziness or neglect. They were economically impossible at $0.039/token human rates. At $0.013/token AI-augmented rates, the math flips.

Look at the cost-per-token shift. Without AI, your output costs $0.045 per token. The we'll get to it backlog isn't laziness — it's rational economics. Nobody's going to spend $0.039 tokens on "nice to have" work when there's $0.039 work that's mandatory.

With AI, your effective cost drops to $0.015 per token. Suddenly the "nice to have" work is cheaper than the mandatory work used to be. The economics invert. NOT doing the extra work is now more expensive than doing it, because your competitors are doing it at the new token rate.

Section V: The $20 Pill

Let's make the NZT metaphor literal. In the movie, a single NZT pill costs $800 on the black market. In reality, the pill costs $20/month. And the math is so absurd it shouldn't be legal.

A 0.2% increase in cost. A 200-300% increase in output. A force multiplier measured in the thousands. If any other business investment delivered this ROI, the board would approve it before the slide finished loading.

And yet. 45-55% of knowledge workers haven't picked up the pill. It's $20. It's sitting on the table. Some are waiting for permission. Some are waiting for training. Some tried it once, got a weird result, and decided it "doesn't work."

In Limitless, there's a scene where Eddie's ex-wife Melissa warns him: "You don't know what that pill is doing to you." She's right to be cautious. But Eddie's also right to respond: "I know exactly what it's doing. It's making me me."

Here's what the token audit reveals: you were never underperforming. You were underequipped. You were producing 210,000 tokens a month through emails, meetings, documents, and analysis — with a 4K context window and a brain that can only hold 7 things at once. That's not a human limitation story. That's a heroism story. You were doing software engineer work on a 4-cylinder engine.

AI doesn't replace the engine. It bolts on a turbocharger. The same driver, the same road knowledge, the same reflexes — just more power. The test harness, the compliance review, the scenario model, the patient education, the real analysis — those weren't aspirational. They were blocked. Blocked by the economics of a $0.039/token human doing $0.039/token work with no room left over.

The $20/month pill doesn't make you smarter. It makes you fully yourself — the software engineer who always knew what the right work looked like but couldn't find the hours.

NZT-48 was fictional. The $20/month subscription is not. The pill is on the table. The question is the same one Eddie Morra faced in that apartment:

Are you going to take it?