Blog

A few weeks ago, Ramp‘s Chief Product Officer Geoff Charles published a thread on X that should be required reading for anyone trying to understand where work is headed. The numbers are arresting: 99.5% of Ramp’s roughly 1,500 employees are active on AI tools. AI usage is up 6,300% year over year. Eighty-four percent use coding agents weekly. Non-engineers (people in sales, finance, legal, marketing) now generate 12% of human-initiated pull requests against the production codebase. Not toy projects. Production code.

The instinct, especially in business education, is to read this as a story about Ramp. A unique culture. A specific tech stack. An anomaly. That reading misses the point. The deeper insight in Charles’s playbook is that access is not adoption. Most companies that hand their employees an AI license and call it a day end up with a fraction of the leverage Ramp is generating, and they’re confused about why.

What Ramp built is the clearest demonstration to date of a shift I’ve been arguing for in my classes for the past two years: the move from analyst to builder. And the implication for how we educate the next generation of business leaders is more urgent than most of us are acting on.

The shift, restated

For most of the modern business era, the dominant professional skill has been analysis. Read the brief. Synthesize the inputs. Frame the recommendation. Hand it off to someone who can execute. MBAs are trained for this. Consulting is built on it. Most of corporate America runs on it.

AI breaks that model. Not because analysis stops mattering (it does), but because the marginal cost of analysis is collapsing toward zero. When a non-engineer at Ramp can stand up a working contract reviewer in an afternoon, the bottleneck is no longer “who can think clearly about this problem.” The bottleneck is “who can get something working in front of a customer or colleague by Friday.”

That’s the analyst-to-builder shift. And Ramp is living proof that when you remove the friction, building isn’t reserved for engineers. It’s available to anyone willing to develop a different set of habits.

Why the Ramp story validates our meta-skills framework

In the Bridge Program and across our MBA work, I teach three meta-skills as the core of what it takes to thrive in an AI-augmented economy:

• Deep Curiosity. Not the surface kind that knows AI is important. The kind that asks what question is actually worth pursuing — and then keeps pulling on the thread when the first answer is unsatisfying.
• Comfort with Ambiguity. The willingness to start a project without knowing the full solution, to sit with a half-built thing, to let the work teach you what it should become.
• Bias to Execution. The instinct to move from analysis to building, fast. To prefer a working v1 over a polished plan.

Source: Geoff Charles [@geoffintech]. “How to get your company AI pilled .” X, April 8, 2026, 6:12pm, https://x.com/geoffintech/status/2042002590758572377?s=20

Read Charles’s playbook with those three in mind and the alignment is striking. His proficiency ladder — L0 through L3 — is essentially a map of how those meta-skills compound over time. L0 is the absence of all three. L1 is curiosity without execution. L2 is the moment someone develops enough comfort with ambiguity to ship something that works. L3 is curiosity, ambiguity tolerance, and execution operating as a system.

The Ramp employees doing remarkable work aren’t the ones who took the best training. They’re the ones who got to what Charles calls the “aha moment” fastest, installed the tool, built something small, and felt the leverage on day one. That’s the threshold that matters. If a tool doesn’t deliver a real result in the first ten minutes of use, most people abandon it, regardless of how much potential it has. Everything that compounds at Ramp compounds from that first concrete win.

What this means for business education

For decades, MBAs carried a reputation as the people who analyzed the problem and recommended the action — and then handed it off. That reputation isn’t entirely fair, and it’s becoming less accurate by the semester. At Tuck, professors like Scott Anthony in Strategy now give students the option to write a traditional final paper or build a working prototype, app, or agent. The number of students choosing to build has grown sharply, and that’s worth celebrating. Our students want to build. The Center for Digital Strategies is leaning hard into that shift, and we’re not the only ones.

Analysis still matters. The ability to frame a problem clearly, weigh trade-offs, and reason from first principles is more valuable in the age of AI, not less. The shift isn’t analysis versus building. It’s analysis as the floor, with building as the new expectation on top.

The companies struggling right now are the ones that didn’t make this transition. A lot of the workforce reductions getting attributed to AI aren’t really about AI doing the work. They’re about organizations carrying too many people who only analyze — who frame problems and recommend solutions but never own execution. Ramp’s playbook is the inverse of that org: everyone is expected to build, and the people who do are compounding their leverage week over week. That gap, between organizations like Ramp and organizations still optimized for hand-offs, is going to define the next decade of careers.

Stanford research is already showing AI’s disproportionate impact on entry-level knowledge workers. The 22-year-old who can analyze but not build is the most exposed worker in the economy. The 45-year-old executive who delegates all building to others is closer to that risk than they realize.

A few specific things this should change about how we educate:

Hiring screens for AI proficiency aren’t optional anymore. Charles describes Ramp’s PM interview: build me a product, walk me through how you built it. Not a slide deck. An actual prototype. Tuck and every peer institution should be asking whether our students can pass that interview before they leave campus, because the companies they’re targeting increasingly will.

The harness matters more than the model. This is Charles’s most important insight, and it’s the one most educators are getting wrong. We spend a lot of time teaching frontier model capabilities and almost no time teaching people how to set up an environment where those capabilities are actually accessible. The MBA student who can quote benchmark scores but can’t get past npm install is going to plateau at L1 forever. The Claude setup mini-lesson I now run before my AI prototyping workshop is, in retrospect, the most leveraged 30 minutes I teach.

The L0 floor is rising fast. A year ago, occasional ChatGPT use put you somewhere in the middle of the workforce distribution. Today it puts you near the bottom. A year from now it will be disqualifying. Curricula built on last year’s baseline are already obsolete.

Building has to be in the room. You don’t develop bias to execution by reading about other people’s execution. The most important pedagogical move I’ve made in the past year is shrinking the time between concept and prototype to the smallest interval the room can tolerate. Forty-five minutes is enough. Two hours is plenty. The mistake is thinking students need a semester to be ready. None of this dismisses the legitimate questions faculty are working through about academic integrity and when AI use supports learning versus shortcuts it. Those questions matter, and the answer isn’t to keep building out of the classroom until they’re resolved. It’s to bring building in and let the pedagogy evolve alongside it.

What I’m taking away from Ramp’s AI approach

A few specific moves from Charles’s playbook that I’m thinking about adapting at CDS.

The skills marketplace concept — Ramp’s “Dojo,” with 350+ shared workflows — is a model for how a CDS Fellows cohort could compound their learning rather than each Fellow figuring it out alone. The first person to crack a workflow shouldn’t be the only one who benefits.

The discovery-through-emulation dynamic Charles describes is one we’re already building this spring. Lily McCarthy T’26 and Clara Delgado T’26, two of our CDS MBA Fellows, are leading a showcase of projects that Tuck students have built this year with the help of AI. The point isn’t to rank anyone. It’s to solve the blank-page problem — the moment a student opens a fresh AI tool with no shared examples, no peer workflows, and no sense of what’s actually possible from where they sit. Making the building visible reframes the question from “do I have to do this” to “what can I build.” Lily and Clara are doing exactly the kind of student-led work that compounds across a community, and I’m grateful for their leadership on it.

Shared context is the other half of the blank-page problem. I recently wrote about a four-part framework for how organizations should build the context documents that make AI genuinely useful inside their walls; the same logic applies to a school. We’re working this term on a baseline context document workshop for Tuck students that captures the frameworks they’re learning, the resources they have access to, and the way we talk about problems in the Center for Digital Strategies. The goal isn’t to hand students a finished artifact. It’s to give them a starting point and then teach them to extend it for their own work. Building a context document is itself a builder skill: it forces you to make tacit knowledge explicit, which is half the battle in any serious project. A student who can articulate what an AI needs to know to be useful in a strategy case has already done the hardest part of the strategy work.

The token budget question is an important question for educators and one I wouldn’t dismiss out of hand. Tuck doesn’t have an unlimited budget, and neither does any university or business school I know of — full token-maxing in the Ramp sense would bankrupt us many times over. What we can do is allocate meaningful token access to students doing serious AI work, treat it as an investment in their development rather than a cost to be minimized, and watch the spend carefully as use scales. The reframe Charles offers, that token consumption is rounding error against the value of a more capable person, is the right principle. The execution has to fit the institution.

The simplest lesson: just get started!

Charles ends his playbook with what he calls the simplest lesson: just get started. That’s the right note to end on here too.

Ramp didn’t have a master plan. They had a culture that valued speed, leadership willing to back bold bets, and a relentless focus on getting people to their first real result. The compounding did the rest.

Business education has the same opportunity. We don’t need a perfect curriculum. We need to stop teaching analysis as the terminal skill and start treating building as the baseline. The students and executives who develop those three meta-skills: deep curiosity, comfort with ambiguity, and bias to execution, will be the ones writing the next decade’s playbook.

The rest will be reading about it.

Patrick Wheeler is Executive Director of the Center for Digital Strategies at the Tuck School of Business at Dartmouth.