Limitless, Without the Pill

For years, I was fascinated by the idea that the human brain contained vast reserves of unused potential. Books on lateral thinking and Renaissance intelligence made me wonder whether a person could become more creative, analytical, technical, and strategic at the same time. Then Limitless turned that wish into a cinematic fantasy: one pill, and every dormant capability suddenly became available.

Going AI-native has felt like the closest real-world version of that fantasy… without the pill, the blackouts, or the Russian loan sharks.

Over the past 60 days, while building Orionfold across research, software, products, publishing, design, and operations, I have discovered that AI does not simply help me do more work. It helps me access different modes of thinking, preserve context between them, and combine capabilities that previously lived in separate departments… or separate versions of myself.

This is not a story about replacing humans with agents. It is about what happens when humans learn to conduct them.

It is about becoming more curious, more multidisciplinary, more ambitious, and perhaps a little more Renaissance in how we work.

The future of AI-native humans may not be artificial superintelligence.

It may be ordinary people becoming far more capable versions of themselves.

How going AI-native unlocked different parts of my brain

For years, I wanted a better brain.

Not necessarily a smarter brain. Just one that could remember everything, connect unrelated ideas, switch effortlessly between disciplines, and finish what it started.

A modest request.

At the time, three works shaped how I thought about this.

Edward de Bono’s Lateral Thinking taught me that the obvious path is often merely the path our brain has used before.

Michael J. Gelb’s How to Think Like Leonardo da Vinci made me wonder whether curiosity, art, science, physical awareness, experimentation, and systems thinking could coexist inside one ordinary human.

Then I watched Limitless.

Bradley Cooper swallowed a transparent pill and went from blocked writer to financial savant, social phenomenon, political force, and owner of several excellent suits.

Naturally, I thought:

Where do I get one?

No productivity app. No morning routine. No second brain composed of 4,700 untagged Notion pages.

Just one pill.

Ideally FDA-approved.

Unfortunately, NZT-48 had several disadvantages: dependency, blackouts, criminals, and a surprisingly high probability of being chased through Manhattan.

AI has so far offered a better risk-reward profile.

Over the past 60 days, I have been building Orionfold as an AI-native business. One person, one desk, multiple products, multiple functions, multiple agents, several local models, and no traditional team.

Something unexpected has happened.

AI has not made me feel as though I am using more of my brain.

It has made me feel as though I can finally use different parts of my brain together.

That distinction matters.

I did not become smarter overnight.

I became more orchestrated.

The 10% dream

In Limitless, Eddie Morra hears a version of the familiar claim that humans access only a fraction of their brains. The film uses “20%,” although popular culture usually rounds it down to 10%. (Wikipedia)

It is a wonderful premise.

It is also wrong.

Modern neuroscience does not support the idea that 80% or 90% of the brain sits around waiting for activation. We use the entire brain over time, including while sleeping. The brain represents roughly 2% of body weight but consumes about 20% of the body’s energy. Nature is unlikely to maintain that much expensive biological equipment purely for decorative purposes. (MIT McGovern Institute)

But the myth survived because it expressed something emotionally true.

Most of us feel underused.

We know we can be strategic, creative, analytical, empathetic, persuasive, technical, playful, disciplined, and courageous.

Just not on the same Tuesday.

At work, we are usually rewarded for narrowing ourselves.

The engineer becomes more technical.

The marketer becomes more marketable.

The executive becomes more executive.

The writer learns to write while someone else handles research, design, distribution, analytics, customer feedback, and the small matter of keeping the company solvent.

Specialization created enormous economic value.

It also encouraged us to leave entire modes of thinking at home.

The real limitation was never that 90% of our neurons were asleep.

It was that our attention, time, tools, and organizations made it prohibitively expensive to activate many capabilities together.

We did not have unused brains. We had poorly routed work.

Digging a better hole

Edward de Bono described traditional reasoning as vertical thinking: logical, sequential, and increasingly deep.

It is excellent when you are digging in the correct place.

The problem is that greater effort does not rescue a bad starting point.

De Bono’s famous formulation was:

“You cannot dig a hole in a different place by digging the same hole deeper.” (IxDF - Interaction Design Foundation)

Lateral thinking asks us to move sideways.

Introduce a random input.

Reverse an assumption.

Use provocation.

Find a new entry point into the problem.

De Bono distinguished between generating the strange idea and producing movement from it. A provocation was not valuable because it sounded clever. It was valuable if it helped the mind travel somewhere useful. (debonogroup.com)

This is remarkably close to how I now use AI agents.

I do not ask one model to give me “the answer.”

I ask different agents to enter the problem from different directions.

One acts as a product strategist.

One behaves like a skeptical customer.

One researches technical feasibility.

One looks for positioning.

One attacks the economics.

One turns the result into a prototype.

Another tries to break it.

The agents are not miniature employees living inside my computer.

They are alternative paths through the problem.

The value is not merely that they work faster.

The value is that they prevent me from digging the same hole with greater enthusiasm.

Leonardo did not stay in his lane

Then there was Leonardo.

How to Think Like Leonardo da Vinci presents seven principles inspired by Leonardo’s life and notebooks: relentless curiosity, learning through experience, refining the senses, embracing uncertainty, balancing art and science, cultivating the body, and recognizing connections between systems. (PenguinRandomhouse.com)

It is difficult to read this without feeling slightly inadequate.

Leonardo painted the Mona Lisa, studied anatomy, designed machines, investigated water, worked on military engineering, filled notebooks with mirrored writing, and still found time to make the rest of us look undercommitted.

The book’s author, Michael Gelb, was himself a professional juggler who performed with the Rolling Stones and Bob Dylan before applying juggling and aikido to learning and leadership. (PenguinRandomhouse.com)

This is either proof of integrated intelligence or evidence that career advice used to be much more interesting.

What fascinated me about Leonardo was not simply that he knew many things.

It was that he appeared to see through disciplines.

Anatomy informed art.

Observation informed engineering.

Water became both physical system and visual form.

Questions migrated from one notebook page to another until boundaries between subjects became less important than the underlying patterns connecting them.

Most modern knowledge work operates in reverse.

We separate research from building.

Building from marketing.

Marketing from sales.

Sales from customer success.

Customer success from product strategy.

Product strategy from financial planning.

We then create meetings so these separated functions can explain themselves to one another.

The Renaissance model integrated the person.

The industrial model specialized the organization.

The AI-native model may allow us to combine the strengths of both.

What happened when I left the organization

At the end of May, I opened Orionfold to the public.

The premise was simple:

One person. One desk. Out in the open.

After nearly nine years at Amazon, I wanted to understand how much useful work one person could produce when the business was designed around AI from the beginning—not retrofitted with a chatbot after the org chart had already hardened. (Meet the builder)

I was not trying to create an automated company in which machines make every decision.

I was trying to create a company in which intelligence can be routed differently.

Orionfold now operates across several connected layers.

Arena is where ideas and models are tested.

Proof turns claims into reproducible evidence.

Relay turns proven capabilities into client workflows with visible cost, margin, and human approval.

Around them sit research, products, developer tools, models, applications, books, websites, and a growing body of field notes. (Explore Orionfold)

In a traditional company, each of these might require a team.

At Orionfold, they began as conversations between me and specialized agents.

One agent worked on strategy.

One worked on marketing.

One worked on the website.

One operated as a research engineer on my local NVIDIA DGX Spark.

I remained responsible for direction, judgment, taste, risk, and deciding what deserved to exist.

The agents expanded the surface area I could explore.

Within one weekend, a paid Arena Field Edition originally planned as a ten-week project was built and shipped. An Advisor product moved from idea to a publicly tested product in roughly 30 hours. One toolbox went through 16 versions in six days. Between March and June, the codebase accumulated around 570 recorded changes. (Read “The Lab That Shipped Itself”)

My public build records now include dozens of field notes, more than 176,000 words, over 63,000 lines in the fieldkit codebase, multiple local models, three books, production websites, applications, benchmarks, and reusable tools. (Read “The Glue Tax”)

These numbers are not presented as evidence that every line is brilliant.

Some of the lines are almost certainly plotting against me.

The point is that one person can now move among research, engineering, product, publishing, design, operations, and distribution without waiting for every function to become separately staffed.

I have a phrase for this:

I do not write every note anymore. I conduct.

The orchestra inside one person

The conductor metaphor can sound grandiose until you understand what is actually being conducted.

It is not an orchestra of artificial geniuses.

It is an orchestra of partially reliable capabilities.

Research agents are good at breadth but need source discipline.

Coding agents can move quickly but occasionally construct elegant solutions to problems nobody has.

Marketing agents can produce twenty headlines in seconds, including nineteen that sound as though a software company has discovered fire.

Analytical agents are excellent at producing tables.

They are less excellent at knowing whether the table matters.

The human role does not disappear.

It moves upward and inward.

I spend less time generating every intermediate artifact and more time asking:

• Is this a real problem?
• Is the proposed solution coherent?
• What would make it useful?
• What feels dishonest?
• What is missing?
• What should be tested?
• Is this good enough to ship?
• Does it belong in the world?

These are not leftover tasks after automation.

They are the work.

AI reduces the cost of expressing an idea through multiple forms.

An intuition can become research.

Research can become an architecture.

The architecture can become code.

The code can become a test.

The test can become a receipt.

The receipt can become a story.

The story can become a product.

The product can generate new questions.

That loop once required a sequence of departments.

Now it can occur inside one person’s working day.

AI does not remove the need for a capable human. It makes human capability more composable.

My brain did not expand. Its switching costs collapsed.

Before AI-native work, changing disciplines was expensive.

To move from strategy to software, I had to load the technical context.

To move from software to storytelling, I had to reconstruct the customer narrative.

To move from storytelling to analysis, I had to find the data, rebuild the assumptions, and remember why any of this mattered.

Every transition incurred cognitive tax.

By the time the brain had loaded the correct application, the day was over.

Agents change this because context can remain active outside my immediate attention.

A research agent can preserve its sources.

A coding agent can retain the architecture.

A marketing agent can remember positioning.

A testing agent can maintain the evaluation harness.

A project agent can record decisions, open questions, and next steps.

I can return to a function without rebuilding it from zero.

This is the closest I have come to the Limitless fantasy.

Not instant genius.

Not perfect recall.

Not the ability to learn Italian during a taxi ride.

Something more useful:

Continuity across modes of thought.

The AI-pilled edge is becoming visible

There is now a small group of people who have moved beyond “using AI.”

They are reorganizing their work around it.

They run several agents at once.

They package repeated instructions into skills.

They maintain persistent context.

They connect agents to repositories, browsers, documents, calendars, messaging systems, and local machines.

They do not merely prompt.

They operate systems.

OpenClaw became a visible symbol of this shift: an open-source, locally controlled assistant capable of working through familiar chat interfaces and performing tasks across a user’s own devices. What began as a weekend project went viral, accumulated hundreds of thousands of GitHub stars according to its official site, and ultimately moved into a foundation-backed open-source model after creator Peter Steinberger joined OpenAI. (GitHub)

The interesting part is not the star count.

It is what people wanted.

They did not want another empty chat window.

They wanted an agent that remained available, knew the environment, and could take action.

The same pattern is appearing inside companies.

OpenAI reports that Codex is used across its departments, including legal, finance, recruiting, engineering, and research. In its 2026 analysis, active Codex usage grew more than fivefold during the first half of the year; more than 10% of users were managing at least three concurrent agents weekly, and long-running tasks increased sharply. (OpenAI)

Early AI-native startups are also testing flatter operating models. Some are serving large user bases with extremely small teams, while research cited by The Wall Street Journal suggests AI-centric companies may reach similar valuations with fewer employees than traditional peers. (The Wall Street Journal)

This does not mean every company will consist of one founder and a warm laptop.

It means the minimum efficient size of an organization may be falling.

More importantly, the minimum efficient size of an ambition may be falling.

Projects that once required permission, capital, hiring, and coordination can increasingly begin with one motivated person.

That is the truly disruptive part.

The FOMO is wrong

The current conversation around AI oscillates between two emotional states.

The first is FOMO:

Everyone else has automated their company.

Their agents wake up at 4 a.m., identify a market, build a SaaS product, negotiate cloud credits, publish a launch video, and send the founder a motivational summary before breakfast.

You are late.

The second is FUD:

AI is useless.

It hallucinates.

The economics do not work.

It will destroy jobs, creativity, education, privacy, democracy, customer support, and possibly the quality of restaurant recommendations.

Both positions contain fragments of truth.

Neither is a useful operating model.

The evidence does not suggest that everyone has figured this out.

Macroeconomic productivity gains remain uneven. Recent reporting indicates daily intensive usage is still concentrated among a minority of workers and businesses. Academic evaluations of long-horizon agents show that even leading systems complete only a minority of complex cross-application tasks reliably. (Financial Times)

This is not evidence that the opportunity is over.

It is evidence that the opportunity has barely begun.

You are not late.

Most organizations are still deciding whether employees are allowed to paste a paragraph into a chatbot.

The frontier is not crowded.

It is mostly people wiring things together and discovering which cable starts the fire alarm.

The FUD is also wrong

The strongest argument against AI-native work is not that the models are incapable.

It is that they are capable enough to create damage while still being unreliable.

That is a serious concern.

Agents can expose sensitive information.

They can take incorrect actions.

They can optimize the stated goal while violating the unstated one.

Persistent agents connected to browsers, messages, files, or payment systems create a much larger security surface than an isolated chatbot. Research into autonomous-agent safety repeatedly emphasizes the need for restricted permissions, isolation, monitoring, and human review. (arXiv)

The response should not be blind optimism.

It should be better engineering.

At Orionfold, I have increasingly organized important work around frozen tests, repeatable inputs, model and configuration identifiers, cost records, explicit approvals, and outputs that can be inspected.

One experiment uses a short configuration hash so a run can be reproduced with the same input and setup. Another local 4B model scored 18 out of 21 on a governed evaluation and refused all nine trick questions. (Read “Same Input, Same Receipt”)

The system has also found its own embarrassing mistakes.

In one case, a paid cloud call was presented as though it were free.

In another, a careful refusal was incorrectly classified as a data leak.

Both bugs were corrected and the tests rerun. (Read “The Fix That Changed the Leaderboard”)

This is what optimism should look like.

Not “the model is always right.”

Not “the model will be right next quarter.”

But:

We can build systems in which being wrong becomes visible, correctable, and less likely to recur.

That is how every serious technology matures.

From knowledge worker to Renaissance operator

The industrial knowledge worker is defined by a function.

The AI-native operator is increasingly defined by an outcome.

This difference changes how we understand capability.

A founder may spend the morning on architecture, the afternoon on positioning, and the evening reviewing an evaluation.

A lawyer may create a research workflow.

A designer may build a working prototype.

An engineer may test a market before implementing the product.

A domain expert may package years of tacit knowledge into a reusable agent.

This does not make every person Leonardo da Vinci.

Leonardo remains inconveniently difficult to benchmark.

But it does revive an older idea: a person can develop across many domains without treating each interest as a distraction from their official identity.

The Renaissance Man was not limitless because he completed infinite tasks.

He was powerful because knowledge moved freely across categories.

AI-native work makes that movement cheaper.

It allows the analytical brain to consult the creative brain.

The strategic brain to consult the technical brain.

The ambitious brain to consult the skeptical brain.

And occasionally, the enthusiastic founder brain to consult an agent whose sole purpose is to ask:

“Are you sure anyone wants this?”

This agent has prevented several masterpieces.

The Seven R’s of the AI-Native Renaissance

After 60 days of building this way, I have arrived at seven working laws.

They are not commandments.

They are closer to warning labels written after touching the hot surface.

1. Route the work

Do not ask one giant agent to become your entire company.

Separate the modes of thinking.

Use one context for research, another for building, another for criticism, another for testing, and another for communication.

Specialization helps agents for the same reason it helps humans: the role clarifies what good work looks like.

The goal is not to simulate an org chart.

It is to give the problem multiple intelligent entry points.

2. Retain judgment

Delegate production.

Do not delegate responsibility.

The human should retain authority over goals, truth, taste, ethics, risk, and final approval.

Your agent may produce the proposal.

You still own what happens when somebody believes it.

The more capable the system becomes, the more important clear human accountability becomes.

3. Require receipts

A polished answer is not proof.

For meaningful claims, retain the source, test, input, output, model, configuration, cost, and decision.

This is especially important when the result looks exactly like what you hoped to see.

AI produces confidence cheaply.

Receipts make confidence expensive again.

4. Remember what works

Do not restart every conversation from zero.

Capture the decisions.

Store successful instructions.

Turn recurring context into skills, templates, tests, and reusable components.

The most important improvement is not that the next model becomes smarter.

It is that your system becomes less forgetful.

5. Restrict autonomy

Start with the smallest useful loop.

Let the agent research before it publishes.

Draft before it sends.

Recommend before it purchases.

Prepare before it deploys.

Expand autonomy only after the workflow has become observable and reliable.

“Human in the loop” should not mean the human watches helplessly as the loop drives away.

6. Rotate models

Do not build a religion around one provider.

Some tasks need a frontier cloud model.

Some need a fast, inexpensive model.

Some need local execution because the data should not leave the room.

Some need several models so their outputs can be compared.

The best model is not a permanent identity.

It is a routing decision.

7. Reuse success

Every successful one-off should leave an asset behind.

A useful prompt becomes a template.

A template becomes a skill.

A skill becomes a workflow.

A workflow becomes a product.

A product becomes a platform capability.

This is how AI-native work compounds.

You do not merely finish the task.

You improve the machine that finishes future tasks.

Route. Retain. Require. Remember. Restrict. Rotate. Reuse.

These are the Seven R’s of the AI-Native Renaissance.

The system I follow: FOLD

The Seven R’s describe how I think.

The system I use each day is simpler.

I call it the FOLD loop.

F — Frame the outcome

Begin with an observable result.

Not “research the market.”

Instead:

“Identify three customer problems, support each with evidence, and recommend one experiment that can be completed this week.”

A well-framed outcome gives both the human and the agent something to reject.

Ambiguity feels creative until five agents interpret it differently.

O — Orchestrate specialists

Assign distinct roles.

For a new product, I might use:

• A researcher to gather evidence
• A strategist to identify the wedge
• A skeptic to attack the assumptions
• A builder to produce the prototype
• An evaluator to test it
• A storyteller to explain it

The exact roles change.

The principle does not.

Different forms of intelligence should collide before the work reaches the customer.

L — Lock the proof

Before expanding the workflow, capture what makes the result trustworthy.

What was the input?

Which model ran?

What did it cost?

Which test passed?

Where did it fail?

Who approved the output?

Can the run be repeated?

This is where a clever demonstration becomes an operating capability.

D — Distill the win

After the work succeeds, package what worked.

Update the instructions.

Save the rubric.

Create the reusable component.

Record the failure mode.

Convert the workflow into something the next agent—and the future you—can use without rediscovering it.

Then fold the result into the next loop.

That is the operating idea behind Orionfold.

Each successful piece of work should make the next piece easier.

A five-day way to begin

You do not need to redesign your company tomorrow.

Choose one recurring piece of work that currently consumes two to four hours.

On day one, write down how you perform it today.

On day two, ask an agent to complete one bounded part of it.

On day three, define a simple rubric for evaluating the result.

On day four, separate production from review: one agent produces, another critiques, and you approve.

On day five, save the successful instructions, examples, and rubric as a reusable workflow.

Do not begin with your most sensitive data.

Do not begin with financial transfers.

Do not begin by granting an experimental agent access to every message you have written since college.

Begin with something useful, reversible, and slightly boring.

Boring workflows are excellent teachers.

They do not generate viral demos.

They generate saved Tuesdays.

What “Limitless” means now

I no longer want to use 100% of my brain.

That sounds exhausting.

I want to use the correct mode of thinking at the correct moment, supported by systems that preserve context, generate alternatives, test claims, and turn successful work into reusable capability.

That is what going AI-native has begun to unlock for me.

Not intelligence from nowhere.

Not instant expertise.

Not a pharmaceutical montage.

A more connected version of the person already there.

The strategist can work with the engineer.

The engineer can work with the writer.

The writer can work with the researcher.

The researcher can work with the entrepreneur.

The entrepreneur can work with the critic.

And all of them can finally leave notes for one another.

The future of AI-native humans will not be humans doing nothing while machines run the world.

It will be humans with greater agency.

A teacher who can create a personalized learning system.

A consultant who can turn expertise into software.

A scientist who can explore more hypotheses.

A small business owner who can finally afford capabilities previously reserved for large companies.

A founder who can test ten ideas before hiring for one.

A curious person who no longer has to choose a single lane before being allowed to begin.

Some people will use AI to produce more noise.

Others will use it to create more leverage, more independence, more experimentation, and more ambitious forms of work.

The distinction will not come from access to the model.

We will all have models.

It will come from the systems we build around them—and the judgment we retain within ourselves.

I once wished for a pill that would unlock the rest of my brain.

It turns out I did not need another 90%.

I needed a better way to coordinate the 100% I already had.

No NZT.

No Russian loan shark.

No side effects observed so far, other than an unreasonable number of repositories.

I am publishing the experiments, products, receipts, failures, and next folds as I build them in Orionfold Stories.

You do not need to follow my exact path.

Start with one useful loop.

Route the work.

Retain the judgment.

Require the receipt.

Then see which part of you wakes up next.