‘The link is not there’: Uber maxxed tokens, not advantage

Uber just burned through its entire 2026 AI budget in the first four months of the year.

Less than 150 days. Hundreds of millions of dollars drained. Just months after initiating a massive push for AI adoption in 2025.

Now, Uber is going “back to the drawing board,” their CTO says, and reassessing how much they spend on AI.

From the outside, the mechanics of how this happened is no surprise. 

To me, the bigger concern is why all that investment produced so little new advantage. 

In late 2025, I watched as Uber made an epic commitment to AI adoption, just like global companies of every kind, tech and not.

On the surface, it worked. 

Since rolling out agentic coding to its 5,000 engineers in December, 95% of developers now use AI tools monthly. An internal coding agent writes roughly 1,800 code changes a week. 70% of committed code is AI-generated. Uber even ranked engineers on a leaderboard measuring their AI tool usage. 

But did anything real come of it?

Uber’s own COO and President, Andrew Macdonald, could not say yes. He admitted as much, publicly.

"That link is not there yet,” he said on a recent podcast. “Maybe implicitly there's more that is getting shipped, but it's very hard to draw a line between one of those stats and 'Okay, now we're actually producing like 25% more useful consumer features.'"

Let that admission sink in: “The link is not there yet.”

The COO and President of one of the most AI-invested companies on earth could not draw a line from that spend to new value. Not that there was none. That he could not yet prove it. 

We already know it’s not an adoption problem. It’s not a technology problem. And it’s certainly not a money problem. 

This is a mindset problem.

There is a clear gap between the companies that are extracting value from AI and the companies that are not.

Uber deployed the mechanics before they had the mindset. 

Uber poured all of its investment into efficiency and execution. 1,800 AI-generated code changes a week. 70% of committed code AI-generated. Engineers graded by usage.

They only aimed AI at the Floor.

Companies that only aim AI at execution, like Uber, will never be able to harvest any new advantage from the technology. They will never free anyone to step onto the Canvas. They will never earn a place in Tomorrow’s regenerative economy. 

You cannot hand a tool like Claude Code to people who have never been asked what new advantage to create, and expect new advantage back. You get a faster execution machine. You get a leaner Floor. You do not get the Canvas.

Mindset must come first. 

The organizations capturing AI’s value are already putting this into practice. The organizations generating an ROI from AI built the mindset before anything else.

The leading companies in the study earned roughly 7.2 times the AI-driven financial gains of their peers.

That concentration of value makes one thing clear: every organization has an urgent choice to make in how it deploys AI.

The choice is not binary. It is not a choice between either using AI to extract more from yesterday's advantage or using it to create Tomorrow's.

The choice is to do both. 

The strongest companies harvest every bit of yield from today’s advantage, and they drive that yield toward regenerating new advantage for themselves, before a competitor beats them to it.

The way I read it, the top 20% of companies in the study did not pick one engine. They refused to choose between execution and regeneration.

They ran both engines at once. They understood the Floor and the Canvas are not a trade-off. They feed each other.

Instead of only pointing AI at efficiency, they pointed it at growth, too. Redesigned workflows. New operating models. More decisions made without a human in the loop.

The execution engine harvests today’s value. The regenerative engine creates Tomorrow’s.

Most organizations are running one engine at full throttle and wondering why the other won't start. Uber ran the execution engine so hard it blew the budget before asking what the regenerative engine needed.

Uber is what it looks like when companies react to AI with aggressive proactivity. Licenses, budget, adoption, all of it, fast.

That is the trap. Proactivity alone is not a virtue. Uber was proactive about extraction and going faster on the Floor, but it was not proactive about regeneration.

Maximum spend, maximum speed. Little new progress.

The Uber news reminds me of the old cobra story from colonial India.

The British had too many cobras in Delhi, so they created a bounty. Bring us dead cobras, and we will pay you.

At first, the incentive appeared to work. People brought in dead cobras. The number went up.

Then people started breeding cobras.

I mean, why hunt wild snakes when you can manufacture the thing being rewarded?

When the British discovered what was happening, they ended the bounty. The breeders no longer had any reason to keep the snakes, so they released them. The cobra problem got worse.

Whether the story happened exactly this way is debated, but economists gave the concept a name: the cobra effect.

When you reward the wrong proxy, you get more of the proxy – not necessarily a solution.

The British got more cobras because they paid for cobras. Uber got more code because they paid for code.

They were tokenmaxxing, believing that value would come from maximum usage. It does not.

AI usage went up. Tokens burned faster. Code changes increased. Engineers climbed leaderboards.

But the COO’s admission still hangs over the whole story:

“The link is not there yet.”

Meaning: Uber could point to adoption, code volume, token burn, and tool usage, but not to a clear connection between all that activity and new value for customers or the business.

This is what cognitive compliance looks like.

AI does not create ROI by existing inside the organization and accelerating the current state. 

It creates ROI when leaders aim it at the right work with the right mindset and objectives guiding the deployment.

This is not just a better approach. It is the intelligent one.

So if maxxing tokens is the wrong thing to max, what is the right one? Imagination!

Imagination is the input that actually creates new advantage. And unlike most inputs, it does not run down when you spend it.

The catch is you cannot buy it. You can only choose where to point it. Maxxing tokens without first maxxing imagination is just spending. You burn the budget before anyone has decided what new advantage you are even trying to create. Point imagination first, and the tokens have somewhere worth going.

That is the difference between Uber and the companies pulling ahead of it.

And it has a name: proactive regeneration.

Proactive regeneration.

You harvest yesterday’s advantage for everything it is still worth, while creating the conditions to expand the possibility space, step into it, and be generative once you are there.

You do it before the economy around you forces the issue. You choose to weaponize commoditization and regenerate before the market commoditizes you.

Weaponizing commoditization means you commoditize your own advantage on purpose. You take its full yield while deliberately driving it toward obsolescence yourself, so you control the timing instead of a competitor controlling it for you. Then you put that yield into the next advantage.

Right now, a lot of organizations are choosing one of two extremes: they’re either waiting too long to take action, or they’re acting to the extreme, as in Uber’s case. 

Activity was abundant. The orientation was missing.

That is why “mindset before mechanics” has to be the principle. Before the tools and the leaderboards, an organization has to ask what it is actually trying to create.

Some of these you will not be able to answer yet. That is not a planning failure. It is why you prototype.

What outcome are we trying to achieve?

What new value should exist because of this?

What negative consequences could we create by optimizing for the wrong signal?

Where are we amplifying human intelligence, and where are we inviting cognitive compliance?

Those questions sound obvious until you look at how many companies are skipping them.

I saw a better version of this recently in a conversation with a leadership team at a software company I work with.

When I asked everyone to share how they were using AI, he explained that junior developers on his team did not automatically get access to Claude Code.

Senior developers had access. Junior developers had to wait.

His reasoning was that junior people had not yet built enough judgment from writing code by hand, failing, fixing, learning and seeing what fails in production. 

I don’t think seniority should automatically equal readiness. That is its own flawed assumption.

But the instinct underneath it was right.

And that conversation allows me to give language to something I had been circling for a while:

Maybe people need a License to Claude.

Not a literal certification, but more of a mindset threshold.

Before we hand someone a tool that can amplify their output, have we helped them build the judgment to know what should be amplified?

I’d hope we all sit with that question:

Do I have my own ‘License to Claude’?

Once the mindset is there, you can start designing for proactive regeneration in practical ways.

One simple way to do this is to pair people side by side. 

A senior engineer brings pattern recognition and scars. A junior teammate brings fewer inherited assumptions and the annoying five-year-old energy of asking why things work this way in the first place.

You might bring customer-facing people into the Canvas, too. I’d call this “Canvas Pairing.” 

A salesperson or customer success leader often has live context that never makes it cleanly into the CRM. Pair them with product, engineering, marketing, or operations and let AI become the shared surface where a vague customer wish can become a prototype before it dies in translation.

Then give small teams enough room to work.

Amazon’s two-pizza team rule — the idea that teams should not grow beyond what two pizzas can feed — worked because smaller teams could move with ownership and less coordination drag. 

The AI version of that could look like small teams with context, authority and enough protection from the Floor to step onto the Canvas.

They are incubators with room for imagination. 

This is where the mindset shows itself. You start to treat tokens, time and attention the way an investor treats capital. You make small bets, you back the ones that prove out, and you let the rest go.

Organizations might create low-cost versions of the Prototypes of Possibility they envision. A Prototype of Possibility is a fast, cheap way to make an idea real enough to test, before you commit anything serious to it.

Then, they should validate these prototypes quickly. Decide what deserves more investment. Kill what does not. Learn from all of it.

That is how you start weaponizing commoditization and initiate proactive regeneration. 

You harvest value from today’s advantage, use AI to make the Floor more efficient, and reinvest that yield into Tomorrow’s advantage before someone else forces you to.

There is a fuller discipline underneath this, closer to how a venture investor runs a portfolio of bets. That is a subject I’ll cover in its own edition. 

For now, the shift is simply this: stop asking how much AI you are using, and start asking what you are investing it in.

After burning through their budget this year, Uber is now imposing an AI coding budget: $1,500 per engineer per month, per tool.

I do not think Uber’s story is an AI failure. I think it is a warning.

Do not spend more and generate more output and then wonder why the link to ROI is not there.

Instead, focus on orientation. Affirm what belongs to machines, what belongs on the Canvas and which human capacities AI should help amplify.

I would leave every leadership team with four questions:

Get this right and your organization starts to look less like an assembly line and more like a living, evolving thing. More organism than org chart. That is where this is heading. More on that next time.

Let’s not just react to AI. Let’s regenerate our advantage now, before the market forces the issue.

That’s what I’m watching for Tomorrow.

– Nish

P.S. Notes from the field.

The last couple of weeks took me through some remarkable rooms.

It started on the Greek island of Syros, at an invite-only, off-the-record gathering. Fewer than 10 of us. From finance, venture capital, geopolitics, and macroeconomics.

I came from where I always do: the intersection of the technology, common-sense economics, and human nature. Some of what I’ve been arguing in these newsletter editions got confirmed in that room. For a humble Canadian kid, watching those heads nod was something.

But the better part was the opposite. Some of my own thinking got updated by theirs. Those exchanges are the whole point.

The conversation is the work.

London gave me two more. A fireside with the in-house creative team at ITV that ran 90 minutes on real back-and-forth, and a Tomorrow's Table dinner with CEOs, creative leaders, and product builders.

Both kept getting pulled to the same idea: AI is a blank canvas of possibility, and we are the unlock.

I’m now hearing people align with that idea more and more.

After a working session last week, a CEO said, “This is what we needed. If we had had this eighteen months ago, we would be much further along."

For almost three years, I’ve said the same thing: the unlock in AI is not in the technology. It is in us. Our thinking, our imagination, our creativity, our curiosity.

That’s landing now in a way it did not used to. Because people are now watching the Uber stories pile up, sometimes inside their own walls.

The most powerful technology ever built, aimed at yesterday's thinking, produces no new advantage.

There is a lot of noise right now, surveys turning thumbs down, fear, fatigue. The leaders I work with are not blind to it. They know the change will be seismic.

They are choosing to think bigger anyway, and the more they do, the more it pulls them to the same place: humans will be needed more than ever.

That is where my mission has settled. Helping leaders and their organizations think and act more optimistically and ambitiously.

Not about AI. About their own potential, individual and collective.

Human potential.

What's coming up: