Key Takeaways

  • AI deployment will be services-heavy in the short term: Oswald Nitski, CPO at Mercor, predicts a significant reliance on AI services because companies currently lack the in-house knowledge to deploy models themselves.
  • Long-term shift to self-deployable AI products: As AI knowledge becomes widespread, Nitski believes the market will transition to sophisticated, self-deployable products, rendering dedicated setup teams less necessary.
  • SF's talent market is "pretty brutal": Harry Stebbings pressed Nitski on the San Francisco talent war, confirming it's incredibly difficult to hire and retain top engineers and product people.
  • "Rocket ships" win with agency and ownership: Nitski argues that only high-growth companies offering true agency and ownership—not just perks—can attract and keep top talent in this competitive environment, even over personality fit.
  • Cyber security AI is an ever-shifting target: Unlike other domains, AI for cyber defense will never reach a stable 90% solution because threat actors constantly move the goalposts, demanding perpetually evolving data and models.

The AI Deployment Endgame: Services Now, Products Later

Forget the dream of AI products slotting seamlessly into your stack tomorrow. Oswald Nitski, Mercor's CPO, offered a counter-intuitive view on the immediate future of enterprise AI deployment. His “hot take,” as he put it, is that for the short term, AI will be a services play. Why? Most companies simply don't know how to use it effectively yet.

“I have a bit of a hot take here. I think it's the future for the short term as the knowledge of how to use AI gets disseminated throughout industry,” Nitski explained. This means companies will hire external teams or internal consultants to bridge that knowledge gap, getting AI off the ground. But this phase won't last forever. As expertise spreads, the market will mature, shifting towards increasingly sophisticated, self-deployable products that don’t require an army of setup specialists. He believes, “eventually it will be, and maybe you won't need at that point teams to go and set things up.” The smart money is on building the service layer now while developing the product for later.

The Only Way to Win SF's Brutal Talent War

Harry Stebbings cut straight to it: “Is the talent war in SF as brutal as it seems?” Nitski didn't mince words. “Yeah, it's pretty brutal. It is very difficult to hire. It is difficult to retain.” The challenge, he said, feels harder than before. But there's a specific cheat code for companies that crack it: becoming a "rocket ship."

Being a rocket ship isn't just about valuation; it's about culture and the type of work you offer. Nitski identified the core magnet for top talent: agency and ownership. These are the qualities that transcend perks or even personality traits. “It's easy when you're on a rocket ship,” he stated. The hard part is identifying these traits. “It's really hard to assess agency and ownership in the interview process,” Nitski admitted, suggesting a deeper challenge for founders beyond standard behavioral questions.

The Ever-Shifting Frontier of AI Data

Beyond general enterprise AI, Nitski highlighted a unique domain: cyber security. This isn't just another data problem; it's an adversarial one where the rules are constantly rewritten. Nitski observed, “We see very rapidly increasing demand for cyber defensive capabilities via data and very interesting data types.” The nature of cyber threats means the goalposts for defense are always moving.

This dynamic is why traditional AI approaches, which aim for high accuracy and stable solutions, will never fully work here. Nitski underscored this by saying, “there's never going to be that 90% for security because the goal posts are always going to move.” For founders eyeing this space, it means building adaptable, rapidly evolving systems rather than seeking a definitive, static solution.

What to Do With This

If you're building an AI company, embrace Nitski's short-term services reality. Launch with a robust service offering that helps customers deploy your AI, then aggressively productize that deployment knowledge. Simultaneously, audit your hiring process tomorrow: pull your last three interview scorecards and look for explicit, measurable criteria for agency and ownership. If you don't have them, design a take-home project or a simulated task this week that forces candidates to demonstrate these traits, not just talk about them.