Abe Ankumah is a Technology leader, Builder & Investor focused on AI Infrastructure and Global Technology Ecosystems.
A quote has been making the rounds. Andrej Karpathy reposted it at the end of April: “You can outsource your thinking, but you cannot outsource your understanding.” It has over 45,000 likes. Boardrooms have been quoting it. It’s resonating because leaders sense the asymmetry but haven’t found the language to name it.
Here’s the strategic claim underneath the slogan: Intelligence is becoming abundant, but understanding is becoming scarce. The gap between them is where the durable advantage will live.
That gap is what every CEO formulating an AI strategy should actually be working on, not “What’s our AI stack?” but rather, “Where does advantage migrate to when intelligence becomes abundant, and how do we build for it?”
That’s the question my article is trying to answer.
It would be easier if “intelligence is getting cheap” were as simple a claim as the headlines make it out to be. It’s not.
Per-token prices have collapsed. Gartner forecasts a 90%-plus drop for trillion-parameter models by 2030. But that’s one cost. Per-call costs may rise as models get more complex. Agentic workflows multiply tokens. The underlying inputs—electricity, water, advanced silicon—don’t follow software’s cost curve.
The more durable claim: Intelligence is becoming abundant. The supply of deployable intelligence is unprecedented and rising fast. Not “cheap.” Abundant. And abundance creates the asymmetry.
When intelligence becomes abundant, scarcity shifts.
It shifts to understanding—the operator’s judgment about what to do with intelligence. This involves knowing what to build, what production failure looks like before it shows up, which customer to listen to and what trade-off is real versus theoretical.
Understanding doesn’t scale the way intelligence does. It’s earned through reps, context, time in the actual work and the pattern-matching that takes years to build. An LLM can summarize a meeting in seconds; it cannot tell you whether your vice president of engineering is about to quit. That gap isn’t closing.
This is the operator’s edge, and it’s becoming the scarce resource leaders need to plan around. This isn’t because AI can’t do anything; it’s because of what AI can’t do, and what only the people closest to the work ever could.
There are three places where understanding shows up as a durable advantage:
1. Infrastructure: These are the systems that hold up under abundant intelligence, including resilience, observability, cost discipline and security architecture that can govern agentic workflows. AI doesn’t just create capability; it creates new failure modes. Companies that survive abundant intelligence are the ones building infrastructure for it, not around it.
2. Process: Workflows should be rebuilt around AI as the default substrate, not the latest tool added on. Most enterprise AI initiatives are layering AI on workflows that weren’t designed for it. Speeding up a broken workflow just gets you a faster broken workflow. The winners are doing the harder work—rebuilding from first principles. Agents replace handoffs, not just typing. Decisions get made closer to the data. Reviews compress because the artifacts arrive better.
3. Ecosystems: These are the networks of trust, expertise and operator access that AI can’t replicate. For example, Israeli cybersecurity, Silicon Valley and, increasingly, certain pockets of Africa, India and Southeast Asia where operator density compounds. These are advantage engines that take decades to build and don’t get disrupted by faster models.
The wrong test most leaders are using: “Does it use AI?”
The right test: “Does this require us to rebuild the work, or just speed it up?”
If it’s the first, the advantage may compound. If it’s the second, you’re paying more for AI to do less-than-optimal work faster—running toward your competitors with better-funded engines doing the same wrong thing.
Notice how this aligns with the cost asymmetry from earlier: Per-call cost is falling, but per-outcome cost only improves if the deployed intelligence is buying something durable. Most enterprise AI investment right now is in the second category.
The companies investing thoughtfully in AI look different from the ones racing to deploy it. They’re going AI-native rather than AI-augmented: AI as the default substrate of the work, not the latest tool added on top. Workflows are rebuilt, not extended. Context flows across the organization. It’s the difference between a business with a website and a business built on the internet.
They invest in understanding as an asset: knowledge graphs that compound institutional context, internal data infrastructure that makes operator judgment legible and networks where the people with judgment can actually be reached. They also hire for context, not just capability: operators with reps in the domain who can tell when AI is wrong because they’ve been on the other side of the same decision before.
None of this looks like better tools. It looks like rebuilt systems and the operators who run them.
The strongest objection to this thesis: AI is moving fast. What looks like “understanding” today might be commoditizable in two or three years. Maybe sooner.
Take it seriously. Then realize that the thesis isn’t really about what specifically stays scarce. It’s about the pattern. Intelligence used to be scarce; it’s becoming abundant. Today’s understanding will be commoditized, too, and something else (agency, trust, taste, human accountability) will be what’s left to win on. The leaders who win across cycles aren’t the ones who picked the right scarcity once. They’re the ones with the systems for spotting what’s scarce now and building for it before everyone else catches on.
The strategic prescription doesn’t change if the window turns out to be shorter. It gets more urgent.
The next decade of advantage won’t go to the leaders who deploy AI first. It’ll go to the leaders who understood where the advantage migrated and built for understanding, while everyone else chased intelligence.
The slogan is right: You can outsource your thinking. You cannot outsource your understanding. But it’s not enough to know it. You have to build the infrastructure, redesign the processes and invest in the ecosystems that turn that asymmetry into a durable position.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
