Seeing AGI (7): The Token Divide — Why Unlimited AI Access Is Now a Corporate Survival Imperative
"The efficiency gap between companies is no longer measured in percentages. In the AI Employee era, companies that give every employee unlimited token access will operate at 10x, 20x — even 100x — the speed of those that don't. This isn't a competitive advantage. It's a civilizational divide."
In my previous six articles, I wrote about AGI's arrival, how individuals should adapt, how to build AI-native teams, the emotional transformation of vibe working, why multi-model architectures represent the future, and how to prepare our children for an AI-native world. All of those pieces were, in one way or another, about people — about how humans need to evolve alongside AI. This seventh piece is different. It's about companies. Specifically, it's about a single organizational decision that I believe will determine which companies are still standing — and which are gone — five years from now.
The Question I Keep Getting Asked
Barely a week goes by without a CEO, a founder, or a senior executive pulling me aside — at a conference, over dinner, on a call — and asking some version of the same question: "How do we actually implement AI in our company?"
I understand why they ask it. It feels like the right question. It sounds strategic, thoughtful, responsible. It signals that they're taking AI seriously. And I always appreciate the intent behind it.
But every time I hear it, I feel a quiet unease. Because that question, while well-meaning, reveals a fundamental misunderstanding of where we are. It frames AI as a project to be implemented — something with a beginning, a scope, a rollout plan, a completion date. It assumes there is a careful, methodical path from "where we are today" to "an AI-enabled company," and that the job is to find and execute that path.
The question they should be asking is far simpler — and far more urgent: "Have we given every single employee in our company unlimited access to think, create, and build with AI?"
If the answer is no — and for the vast majority of companies I speak with, the honest answer is no — then everything else is noise. No AI strategy deck, no task force, no pilot program, no governance framework matters until that foundational question is answered. Because without it, you're not implementing AI. You're performing the idea of implementing AI. And the difference between those two things is everything.
What "Unlimited Token Access" Actually Means
Let me get specific, because I think a lot of people hear "unlimited AI access" and imagine it means something vague — something like "a permissive culture around AI tools." It doesn't. It means something precise and measurable.
Tokens are the fundamental unit of AI work. Every query you send, every document you ask an AI to analyze, every piece of code you ask it to write, every agentic task you set in motion — all of it consumes tokens. Tokens are, in the most literal sense, the raw material of AI-powered productivity. They are to the AI era what kilowatt-hours were to the industrial era, what bandwidth was to the internet era.
When a company sets monthly token caps — when it requires employees to submit requests through IT to get access to a frontier model, when it blocks certain AI tools on the corporate network, when it offers one shared team subscription for twenty people, when it mandates that AI usage must be logged and reviewed — it is rationing its employees' ability to think with AI. It is placing a governor on their cognitive output. It is literally limiting how much intelligence they are allowed to apply to their work.
Think about how absurd that sounds when you say it out loud. We are limiting how much intelligence our employees are allowed to use.
And yet, this is what most companies are doing today. Not out of malice, but out of habit — the same instinct that leads IT departments to treat every new technology as a cost to be controlled rather than a capability to be unleashed.
I lived through a similar moment in the early 2000s. Some companies gave every employee unrestricted internet access and said: use this however you need to do your job better. Other companies locked down internet access, blocked sites, monitored usage, and issued corporate policies about what was and wasn't permitted. The companies that gave unlimited internet in 2000 are largely the companies that dominated their industries in 2010. The ones that rationed it didn't lose because their internet policy was bad. They lost because their fundamental orientation toward new technology was wrong. They were optimizing for control when they should have been optimizing for capability.
But there is an even older — and more instructive — parallel, one that I keep coming back to.
In the early 1900s, electric motors became widely available to industrial factories. Most factory owners made what seemed like the obvious move: they replaced their central steam engine with an electric motor. Same belts, same line shafts, same factory layout — just a different power source. The result was modest. Efficiency improved a little. Costs fell somewhat. But the transformation was nothing like what was possible.
The factories that unlocked the real revolution did something entirely different. They threw out the entire transmission system — the line shafts, the belts, the central-power logic — and installed individual electric motors directly at each workstation. Then they redesigned the factory floor from scratch around this new architecture. Output didn't improve by 10% or 20%. It improved by three, four, sometimes five times. Workflows that were previously constrained by physical transmission distance could be reorganized for logic and speed. New production methods became possible that were literally impossible under the old architecture.
The economist Paul David studied this phenomenon in a famous 1990 paper. He called it the "dynamo paradox": electricity had been commercially available for nearly four decades before most factories saw transformational productivity gains — because the majority of them were using new power with old thinking. They had adopted the technology. They had not reorganized around it.
Here is what strikes me most about this: the factories that failed to transform weren't uninformed or negligent. They had access to the same electric motors as their competitors. They were paying for electricity. They genuinely believed they had adopted the new technology. What they had actually done was bolt new power onto an old structure — and then wondered why the results weren't proportional to the investment.
I see the exact same pattern playing out today. Most companies deploying AI are doing the factory equivalent of swapping the steam engine for an electric motor and calling it a transformation. One shared subscription. A few approved use cases. A governance framework. A quarterly AI review meeting. The old organizational structure — the old belts and line shafts — remains completely intact.
Unlimited token access is the organizational equivalent of ripping out the line shafts and putting individual motors at every workstation. It is not merely a cost decision. It is a structural decision — a signal that you are redesigning the factory floor, not just changing the power source. And like those early 20th-century factories, the companies that make this structural choice will not improve by 10%. They will operate in a fundamentally different category of productivity than those that don't.
We are at that same fork in the road today. Except the stakes are incomparably higher — and the gap between the two paths opens incomparably faster.
The AI Employee Era Is Already Here
In my third article, I wrote about building AI-native teams. In my fourth, I described vibe working — the psychological and operational transformation that happens when humans stop treating AI as a tool and start treating it as a genuine collaborator. Those pieces described a transition in progress.
I want to be clear now: that transition is no longer in progress. It is complete. The AI Employee era has arrived.
AI is not a tool you use to write emails faster. It is not a search engine you query when you're stuck. It is a co-worker. A co-founder. An army of specialists — engineers, researchers, analysts, strategists, designers, writers — available to every person in your company, twenty-four hours a day, seven days a week, without vacation, without ego, without organizational politics. AI doesn't go home at six. It doesn't lose motivation. It doesn't need three weeks to onboard. It doesn't require a recurring salary negotiation.
But here is the thing about this army: it only shows up if you open the gates.
At Genspark, we went from zero to $200M in annual run rate in eleven months — a pace that, as far as I know, has never been seen before in enterprise AI. We hit $10M ARR in the first nine days after launch, faster than ChatGPT, faster than Claude, faster than any AI product in history. We did this with a team that was, by any traditional standard, absurdly small for the output we were producing. AI writes 100% of our code. One engineer built our AI browser in three months. A PM delivered AI Slides in two weeks. A designer who had never coded before built a browser download site in three days. In the eleven months since, we've shipped AI Workspace 3.0, Genspark Claw — our first fully autonomous AI employee — plus Workflows, Teams, Meeting Bots, Realtime Voice, and more. These are not exceptional human beings — they are people of normal talent who have been given unlimited access to an exceptional army.
A company with 50 humans and unlimited AI access does not operate like a company of 50. It operates like a company of 500, or 5,000. The multiplier is real, it is measurable, and we live it every single day.
Now consider the other scenario: a company with 500 humans and restricted AI access. Monthly token budgets. IT approval workflows. A cautious pilot with one department. Quarterly reviews. A carefully governed rollout plan.
That company operates like a company of 500. Nothing more.
The 50-person company will ship ten times faster, iterate ten times more, fail ten times more productively, and learn ten times more quickly. And every week that passes, the gap between them grows wider.
The Light-Year Gap: Why This Time Is Different
In every previous technology wave — the PC era, the internet era, the mobile era, the cloud era — there was a gap between early adopters and laggards. Companies that moved faster gained advantages. But those advantages, while real, were bounded. The efficiency gap between an early internet adopter and a late one was maybe 1.5x. Maybe 2x. Perhaps, for the very best-run companies, 3x.
Those gaps were recoverable. A company that was two years behind on cloud adoption in 2012 could catch up by 2015. It was painful and expensive, but possible.
What is happening now is categorically different. The gap is not linear. It is exponential. And I am not sure it is recoverable.
Imagine two ships leaving the same port on the same day. One is powered by nuclear propulsion. The other has oars. On day one, the nuclear ship is a little bit ahead. By the end of the first week, it is far ahead. By the end of the first month, the oar-powered ship cannot even see the nuclear ship on the horizon. By the end of the first year, the distance between them is not large. It is not very large. It is incomprehensible — a gap measured not in miles but in a different category of reality altogether.
That is what the token divide is creating between companies right now.
On one side: a company where every employee has unlimited access to the most powerful frontier AI models — where engineers are having live, multi-turn conversations with AI to architect entire systems, where product managers are generating and iterating on research reports in minutes rather than weeks, where executives are stress-testing strategies against AI-generated competitive scenarios before a single slide deck is built. Every person in that company is operating with a cognitive multiplier that compounds daily.
On the other side: a company where accessing a frontier AI model requires submitting a ticket to IT, where the budget for AI tools is debated quarterly, where employees have found workarounds using their personal credit cards because the official tools are too restricted, where leadership is still deciding whether to expand the pilot from the engineering team to the marketing team.
The difference in output between these two companies is not 10%. It is not 50%. It is the difference between a company that is running and a company that is standing still. And every single day that passes, the running company gets further ahead — not linearly, but exponentially, because faster iteration means faster learning, which means better products, which means more revenue, which means even faster iteration.
This is not a competitive advantage. It is, over time, an extinction event.
What "Fully Embracing" This Actually Looks Like
I want to be concrete, because "fully embrace AI" is the kind of phrase that sounds meaningful but can be used to describe almost anything.
Remove all token caps and spending limits on AI tools for every employee — today. Not next quarter. Not after the security review is complete. Today. Yes, there will be cost. That cost is trivially small compared to the productivity gains, and orders of magnitude smaller than the cost of falling behind.
Stop treating AI as an IT expense. AI spend belongs on the headcount budget — not the IT budget. The moment you move that line item, everything changes. It signals to your entire organization that AI is not a software tool to be managed and minimized — it is a member of the team. Think of it this way: every AI agent your company deploys deserves its own seat, its own workstation, its own reporting line. It has a role. It has deliverables. It has a manager. When you treat AI that way — when it shows up in your org chart, not just your vendor contracts — your people start treating it that way too. No reasonable CFO looks at the salary line and thinks: "How do we cut this to reduce costs?" Salaries are the price of human capability. AI access is the price of AI capability. In a world where AI is doing 80% of the work, that line item deserves the same respect — and the same investment philosophy — as the people sitting next to it.
Create an internal culture where using AI for everything is the default, not the exception. At Genspark, we don't have a policy that says "use AI when it makes sense." We have a culture that says "if you didn't use AI for this, tell us why." That inversion matters enormously. It signals organizational seriousness. It creates peer accountability. And it accelerates collective learning, because when everyone is using AI aggressively, the learnings spread fast.
End the pilot phase. I want to be direct about this: if your company is still in the "testing and evaluating AI" phase, you are not being prudent. You are being slow. The time for pilots ended in 2023. The companies that are winning right now are not piloting — they are deploying, iterating, and compounding. Every month you spend evaluating is a month your competitors spend executing.
The Companies Getting Left Behind Right Now
I want to paint a picture — not to shame anyone, but because I think some leaders genuinely don't realize how the external world sees them.
The company that is falling behind has an AI strategy committee that meets monthly. It has approved one shared AI subscription per team of twenty. It is running a pilot with the engineering team and plans to "expand to other departments" after the first pilot completes — which will take another two quarters. Its employees are using their personal phones and personal credit cards to access frontier AI tools that their employer won't provide, not because they are breaking rules, but because they need to do their jobs.
Leadership in this company believes they are being responsible. They are managing risk. They are moving carefully.
Meanwhile, at another company, a 26-year-old engineer is having a real-time conversation with an AI agent that is simultaneously writing code, running tests, analyzing errors, and suggesting architectural improvements — all in the time it takes the first company's engineer to file a ticket asking for access to a basic AI coding assistant. By the time that ticket is approved, the 26-year-old has shipped a feature.
These two companies are not operating in different eras. They are operating in different civilizations.
The New Corporate Wealth Divide
Here is what concerns me most — and I want to be precise about this, because it's easy to feel the urgency without fully grasping the mechanism.
The token divide is not just a gap in current output. It is a gap in the speed of learning. And that is what makes it so dangerous.
A company that has given every employee unlimited AI access for the past two years has not just done twice as much work. It has accumulated two years of organizational learning — practices, instincts, muscle memory, internal culture — that simply cannot be replicated by writing a check. You cannot acquire your way to AI-native. You cannot hire your way there in six months. Organizational readiness compounds quietly, invisibly, until the gap between the company that built it and the company that didn't is not a performance gap — it is a capability gap of a different order entirely.
The companies that moved first and fully are now inside a flywheel that is almost impossible to stop. Their products are better, so they attract more users. More users generate more data and feedback, so their products get better faster. Faster iteration means faster learning, which funds more AI investment, which accelerates iteration further. Meanwhile, their best talent — the people who thrive in AI-native environments — gravitates toward them, because no ambitious engineer or designer wants to spend their career waiting for IT approval to access a frontier model.
The laggards, by contrast, face a compounding deficit. They are not just behind on output — they are behind on instinct, behind on culture, behind on talent density, and behind on the flywheel itself. And at some point — and this is the part that genuinely frightens me — that deficit crosses a threshold where it is no longer a question of catching up. It is a question of whether you are still in the same race.
You cannot row your way back into a race against a nuclear-powered ship. Three years behind on the token divide may be permanent. That is not a metaphor. I mean it literally.
Final Thoughts
I've spent the last several months watching two kinds of companies. The first kind is moving with the wave — not surfing it perfectly, but moving. They are making decisions quickly, accepting uncertainty, embracing the messiness of full AI deployment, and compounding their learning every week. The second kind is still standing on the shore, watching the wave approach, convening meetings about whether to get in the water.
I wrote my first "Seeing AGI" piece as a father worried about my 12-year-old son's future. I feel that same parental concern now — but directed at the founders and operators reading this. Because I have seen what is coming, and I genuinely don't want anyone to be swept away by it.
When a tsunami hits, it does not wait for you to finish your board meeting. It does not pause while you complete your governance review. It does not give you one more quarter to expand the pilot. It arrives, and the organizations that were in the water — moving with it, working with its energy — survive and advance. The organizations that were still deliberating on the shore get buried.
The window is not closed. But it is closing. And the question every founder, every CEO, every operator reading this needs to answer tonight — not next week, not next quarter — is this: Have you given every single person in your company unlimited access to think, create, and build with AI?
If not, I want you to ask yourself one more question: What are you waiting for?
I've been in technology for nearly twenty years. I've seen markets shift, companies rise and vanish, and paradigms flip overnight. But I have never seen anything move this fast, or cut this deep.
And the thing that keeps me up at night is not the technology itself. It is the image of brilliant, hardworking people — founders who sacrificed everything to build their companies, operators who gave years of their lives to their teams — waking up one day to discover that the gap between them and their competitors is no longer a gap they can close. Not because they weren't smart enough. Not because they didn't care enough. But because, at a critical moment, they hesitated. They waited for one more data point. They convened one more committee. They asked for one more quarter to evaluate.
I don't write these articles to be alarmist. I write them because I genuinely believe most people haven't felt the full weight of what is happening yet — and by the time they do, it may be too late to act.
So let me leave you with the one thing I most want you to carry from this piece.
The efficiency gap between companies is no longer a matter of talent, strategy, or capital. It is increasingly a matter of one decision: have you given every person in your company unlimited access to think, create, and build with AI — or have you not? The companies that said yes, even imperfectly, even messily, are compounding their advantage every single day. The companies that are still deliberating are not standing still. They are falling behind at a rate that history has no precedent for.
That gap used to be measured in percentages. It is now measured in multiples. Soon, for some industries and some companies, it will not be measurable at all — because one side of the equation will simply no longer be in the race.
I hope you are on the right side of that line. And if you're not sure — if you read this and felt even a flicker of recognition, a quiet voice saying "this might be us" — then please, don't wait for the next board meeting to find out. The wave is already moving. The only question is whether you're in the water or on the shore.
There is still time. But not as much as you think.