George Sivulka argues that managing AI agents well is the new competitive moat, drawing parallels to the railroad era's demand for sophisticated management
George Sivulka, the CEO of AI platform Hebbia, dropped a provocative line on X that cuts to the heart of where enterprise labor is heading: “100X tokens are the new 10X engineers.” The implication is blunt. In a world where human engineers cost roughly $80 per hour, a well-orchestrated fleet of AI agents can accomplish work at a fraction of that cost, or dramatically more, depending entirely on how skillfully they’re managed.
The range he cited is striking. Well-managed agents can cost anywhere from $4 to $7,000 per hour depending on scale and usage. That’s not a typo. The spread between the floor and ceiling tells you everything about why Sivulka thinks “token management” is becoming the defining skill of the next decade.
The railroad analogy and why management matters more than models #
Sivulka framed the argument through a historical lens, comparing the current AI agent moment to the emergence of railroads in the 19th century. Railroads didn’t just need tracks and locomotives. They needed an entirely new class of managers to coordinate schedules, logistics, and capital allocation at unprecedented scale.
AI agents, in his view, present the same challenge. The models themselves are increasingly commoditized. What separates a $4-per-hour deployment from a $7,000-per-hour money pit is the quality of orchestration sitting on top. Only a small fraction of users actually know how to prompt and manage AI agents effectively. That gap between competent and incompetent usage is where the real economic story lives.
Who is Hebbia and why should you care #
Sivulka founded Hebbia in 2020 after leaving a PhD program at Stanford. The company isn’t building chatbots for customer service queues. It’s building agent-based workflows for institutions that manage enormous amounts of data and capital.
The client list reads like a who’s-who of institutional power. BlackRock, KKR, and the US Air Force all use Hebbia’s platform.
Hebbia also has a strategic partnership with OpenAI, focused specifically on finance, banking, and legal automation. A junior analyst at a private equity firm spending 60 hours combing through a data room is exactly the kind of labor these systems are designed to augment or replace.
In a May 2025 blog post published through Radical Ventures, Sivulka laid out his vision for what he calls “agent employees,” essentially AI systems that don’t just answer questions but execute multi-step workflows with the reliability you’d expect from a human team member.
What this means for investors and the broader market #
The core investment thesis here is straightforward. If AI agents genuinely deliver 10x or 100x cost efficiency improvements over human labor in knowledge work, then companies that master token management will have a structural cost advantage over those that don’t.
The $4 to $7,000 range Sivulka cited also suggests something important about market segmentation. Small-scale, well-scoped deployments can be absurdly cheap. Large-scale, poorly managed deployments can be absurdly expensive. That mirrors early cloud computing adoption, where some companies slashed IT costs by 80% and others accidentally racked up six-figure AWS bills because nobody was watching the dashboard.
The companies that will win this cycle aren’t necessarily the ones building the best models. They’re the ones building the best management layers on top of those models. Hebbia is betting its entire business on that premise, and given its client roster, the bet appears to be working.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our