Goldman’s CEO Told This Summer’s Interns to Spar With the AI

Every summer, Goldman Sachs pulls its intern class from an applicant pool so large that fewer than one in a hundred make it in. This year, David Solomon spent part of his time in front of them talking about AI. His advice was to argue with it.

“Ask questions, challenge the conventional wisdom,” he told them. “You have the ability to pick any topic and, in a much shorter period of time, get much smarter on it. But you have to challenge what the models give you.”

Solomon treats LLMs as sparring partners, according to the author. He feeds a model his reading of a situation, then pushes back: What are the implications? What would it take to flip this assumption? He credits the habit to his father, who used to stop young David before a big decision and force him to sit with the option he had already written off. Solomon is running that same playbook on AI, and he wants his interns doing it too.

It is a striking message coming from Wall Street. The CEO of one of the most powerful financial firms in the world told his top young hires that AI’s true value isn’t fast output. It is the friction it creates in your thinking.

For the last several years, corporate tech messaging centered around one core concept: use AI to compress hours, expand productivity, and speed up your work. Some companies cut hiring accordingly. Students watching the shift rushed into asynchronous AI learning to protect their economic viability. Educators (including myself) panicked about the value of our traditional learning and assessment structure. For some, assignments began to feel like hollow formalities. What do we do when leading companies signal to young people that the best thing they can do for their own hiring prospects is to use a tool that allows them to bypass the exact mental struggle that education is meant to build? The impact of this disconnect was then and remains today palpable.

But that dynamic is shifting. Job openings jumped by 731,000 in April, led by professional and business services. Companies that cut early are rebuilding those teams. In a Robert Half survey of hiring managers, a third said they cut a role due to AI only to rehire for it later. In finance, that number hit 44 percent.

The executive narrative has changed along with the numbers. In a New York Times guest essay earlier this year, Solomon acknowledged AI-driven disruption while arguing that human judgment becomes more critical, not less.

“The technology has no ability to do what we do as human beings, to think about whether something makes sense,” he noted. “There’s a form of intelligence which is analyzing data, but there’s also a form of intelligence which is taking life experience and using it to infer things data cannot.”

As the corporate world begins to shift its rhetoric, educators should take notice. The emphasis on a “form of intelligence” that comes from “infer[ring] things that data cannot” indicates a fundamental change in what education has to measure.

When AI handles content retrieval, the value moves towards evaluating the human behind the answer: their judgment, their communication, and their ability to pivot under pressure. The irony is that AI opens up new ways to evaluate those exact human skills at scale. It may just create the solution to the very problem it introduced.

Watch what universities are doing right now and you can see alignment to what Solomon and others are intimating. Faculty are bringing back the oral exam. At Cornell, students defend their written work in twenty-minute Socratic sessions, because, as the professor who runs them puts it, you can't AI your way through an oral exam. At Penn, the head of the teaching center describes a massive shift toward in-person assessment. A professor at NYU Stern built an AI-driven oral exam for one reason: to find out whether a student actually understands the work their team handed in. The common thread is a hunt for a way to watch a person reason in the moment, where the thinking can't be outsourced.

Employers are naming the same target. The World Economic Forum’s Future of Jobs report ranks analytical thinking as the most essential skill in the workforce, called critical by seven of ten companies, with resilience, leadership, and creative thinking close behind. Leadership and social influence climbed twenty-two points in two years. NACE’s 2026 outlook has employers listing critical thinking, communication, and teamwork as the competencies they recruit for, and seventy percent of them now hire on demonstrated skill rather than credentials. A GPA no longer tells that story. The market is moving towards a wholly reinvented manner of assessing human capability.

If you’re an educator, here’s an idea that’s worked for me. Give students a position to defend and let an AI push back on them, then read the conversation that comes out. Ten messages of that has shown me more about how a student thinks than their final essay does. You don’t have to redesign your course. One assignment is enough to see if it’s worth more.

If you’re hiring, the point is probably already top of mind for you. Résumés and cover letters used to tell you something real. Even situational judgment tests on-site are no longer as trustworthy as they once were. These outputs need to be re-assessed in the same ways that educators are reconsidering the essay itself. Focus on assessment that shows how someone thinks on their feet when there’s no clean answer. Finding a way to watch that, even a rough one, will tell you more about who you’re about to hire than any traditional test.

If you’re a student or just starting out, Solomon’s advice is what I would focus on. Your potential future boss is going to be less and less concerned with what you make when you use AI, and a lot more concerned with how you use AI. Come to your interviews ready to describe a way that you used AI to push you, and what happened to your thinking as a result. They’ll find that far more interesting than a slide deck, piece of writing, or even AI agent that you built.

Solomon’s comments are a signal. The good news is that industry and higher ed may be coming together faster than expected.

On August 19th, my partners and I are hosting a live virtual demo of our AI Friction Labs platform. We'll run one of our sparring partner simulations and walk through the tools we use to measure human skills. Then we hand it over and let you go a round with one. We're also opening a free Explorer tier, so you can work through the whole thing the way a student would. Register here.