Axel, a second-year MBA student at the University of Virginia's Darden School of Business, had a practical question. Many of his classmates were heading into consulting and banking, but he was leaning toward tech and entrepreneurship. With things moving so fast, he asked, what was the smartest thing for someone like him to do in the next 12 to 18 months? Should he join a fast-growing AI or hardware company, start something of his own as soon as possible, or do something else entirely?
He asked it during the second audience "Ask Me Anything" session of the Moonshots podcast, hosted by Peter Diamandis, founder of XPRIZE and Singularity University. The episode was recorded on September 18 and published on September 22, 2026. The panel's answers to Axel, and to a later question from a workforce specialist named Adam, kept returning to coordination. One question was how people can direct AI systems. The other was how AI might match people and agents to work. The panel also discussed how long humans would remain useful in that arrangement.
Managing agents like a workforce
Diamandis handed the MBA question to Dave Blundin, founder and general partner of the venture firm Link Ventures. Blundin told Axel to "get out and get going as quickly as you can." He saw "a massive opportunity in management of agents", which he described as wide open for some period of time, "but certainly right now."
An AI agent is software that is given a goal and carries out a series of steps to reach it, often using other programs and tools along the way. When many agents work on the same job, someone has to decide how to split the work and how the agents share results.
Blundin said he was shocked by how much of his experience managing people carries over to that task. Getting a thousand agents to do something constructive together, he said, is "shockingly similar" to getting a thousand people to do the same. He named the parallels: how the workers communicate, the level of detail they use when communicating, and how the problem is divided among them.
Getting into the room where deals happen
Blundin's second piece of advice was about deals, not agents. He said many people go wrong by going "into a basement" and trying to build something. In his view, the "entire restructuring of the world" is happening through a huge amount of dealmaking. He challenged Axel to cut the time between graduation and his first role in a $5 billion or $10 billion negotiation to 30 days or less.
Blundin said that at OpenAI and Anthropic, hundreds of negotiations worth $5 billion to $10 billion each are going on at the same time, and that they are "massively understaffed." He argued that getting into that kind of work within 30 to 60 days of graduating would give Axel the fastest possible start. Only then, he said, should Axel "work back from that position into what you want to build."
Starting from capabilities instead of jobs
Diamandis then brought on Adam, who approached coordination from the organization's side. Adam said he has spent 25 years working on talent and workforce systems for major corporations, universities and governments in the UK, India, China, the Middle East and North America. He said his recent work concerns how humans and AI work together and what might eventually replace today's job-based structure of work.
Adam noted that most thinking about the future of work still starts with demand. Someone spots a problem or an opportunity, and then a mix of human and AI capability is put together to deal with it. He wants to reverse that. Suppose AI systems could keep track of what people and agents can do as their abilities change. How well, he asked, could those systems discover new combinations of talent and resources and automatically match them to new opportunities to create value?
Salim Ismail: build for learning speed
Diamandis called it a question for Salim Ismail, founder of Open ExO and a general partner at Exponential Venture Capital. Ismail told Adam he was "exactly on the right track." He said new AI systems have feedback loops that let them pick up what is happening and learn tacit knowledge very quickly. Tacit knowledge is the practical know-how people rarely write down.
Ismail said organizations should be designed to speed up those learning loops and absorb new lessons as fast as possible. His advice to chief executives, he said, is to invest in their organizations' adaptability and flexibility and "double down on that." He described this as part of his organizational-singularity work, which involves rebuilding AI-centered workflows on top of an "intelligence stack". Ismail defined that stack as a learning loop, with workflows layered on top. The goal, he said, is for the whole system to become "a self-learning proposition."
A window of a few years
Alexander Wissner-Gross, a computer scientist and founder of Reified, added a time limit. Saying "if I understand the question," he answered that AI can usefully and productively direct human activity only during a "likely narrow window" of a few years, "at most 10 years maximum." After that, he said, humans would need to merge with machines for their contributions to remain economically relevant. "I don't think it's an indefinite window," he said. "I think it's a finite window."
Blundin agreed that the window lasts a few years but said that was no reason to hesitate. If people gain "a huge amount of leverage" during that time, he argued, they can branch out from that position.
The Mercor model
Blundin's example was Mercor, which he called "the fastest appreciating company in the history of the world." He advised following in its footsteps and studying everything the company did, looking past the narrow version of its business. According to Blundin, Mercor has "unleashed a hundred thousand people now to help with AI," all of them working as individuals. "How does that work?" he asked. He did not give a source for either the superlative or the workforce figure.
Mercor describes its business in its Series C announcement of October 27, 2025. The company pays professionals to work with AI labs and businesses. Doctors contribute clinical judgment, bankers teach financial analysis, and lawyers help refine reasoning about legal precedent. The company says the work goes beyond ordinary freelance assignments. The experts turn their judgment and working methods into evaluations: graded tests of a model's performance that models can also learn from. In the announcement, Mercor reported raising $350 million at a $10 billion valuation, five times its Series B valuation. It said it would spend the money on growing its talent network, matching experts to work more accurately and delivering faster.
That description of finding out what individual experts know and matching them to the AI projects that need it overlaps with the matching problem Adam raised. It was the kind of model Blundin urged the callers to study while the window he and Wissner-Gross described remains open.