Agentic AI · Enterprise Technology · Startups

OpenAI has backed a startup building software that coordinates thousands of AI agents at once. The technology is real, the valuation is striking, and the implications for enterprise buyers are worth examining carefully.

Shashi Bellamkonda  |  March 27, 2026

The dominant model for artificial intelligence in enterprise settings has been the single powerful model: one system, one prompt, one output. That architecture works well for bounded tasks. It struggles when the problem is genuinely complex, when the data is distributed across many sources, and when the question requires multiple kinds of reasoning running in parallel. A San Francisco startup called Isara is building a direct answer to that constraint, and OpenAI has decided the bet is worth backing at a $650 million valuation.

Isara was founded in June 2025 by two researchers who were 23 years old at the time. Eddie Zhang was completing a doctorate in computer science at Harvard and had previously worked on AI safety research at OpenAI. Henry Gasztowtt was studying computer science at Oxford University. The two had co-authored an academic paper in June 2024 examining how AI systems might work together to improve policymaking. That paper became the conceptual foundation for the company.

Since founding, Isara has hired more than a dozen researchers from Google, Meta Platforms, and OpenAI. The company has raised $94 million from investors including Michael Ovitz and Stanley Druckenmiller, with OpenAI's investment placing the valuation at $650 million. For a company less than a year old, those numbers reflect how much capital is currently chasing the multi-agent systems category.

What Multi-Agent Coordination Actually Means

The term "AI agents" has been used loosely enough in marketing materials that it has started to lose meaning. What Isara is describing is more specific and more technically ambitious than most vendor definitions of the word. The company's approach involves deploying large numbers of smaller, specialized AI agents that communicate with each other, divide tasks, and synthesize their outputs into a coherent result. The vision is closer to a functioning organization of AI workers than to a single AI assistant with a longer task list.

CJ Reim, whose venture firm Amity Ventures backed Isara, described the approach as coordinating swarms of specialists to perform research for users. That framing is useful for understanding what problem the architecture is trying to solve. Individual AI models are generalists operating at high volume. Isara's premise is that specialist agents, coordinated intelligently, can outperform a single generalist on problems that require depth across multiple domains simultaneously.

The company demonstrated roughly 2,000 agents coordinating to forecast the price of gold at a technology conference hosted by Allen and Company in Arizona earlier this year. That is not a product demo in the traditional sense. It is a research proof of concept with significant engineering behind it, designed to show that the coordination layer works at scale before the vertical applications are fully built out.

The Initial Market and the Expansion Path

Isara's stated plan is to initially sell its software to investment firms for predictive modeling, along with other financial services companies. That is a sensible starting point. Financial services organizations have well-defined data problems, clear performance metrics, existing tolerance for algorithmic decision support, and procurement budgets aligned with technology that demonstrably moves numbers. It is a market that can validate the technology without requiring the buyer to fundamentally change how they make decisions.

The longer-term expansion targets are finance broadly, biotech, and geopolitical analysis. Each of those domains shares the same structural characteristic: the underlying questions are too complex and too data-intensive for a single model to handle reliably. Biotech research involves navigating scientific literature, experimental data, regulatory history, and competitive intelligence simultaneously. Geopolitical analysis requires synthesizing economic indicators, historical patterns, political signals, and real-time events across multiple regions. These are exactly the kinds of problems where a coordinated swarm of specialist agents would theoretically outperform a single generalist model.

Whether the coordination layer can deliver on that premise at production scale, across real enterprise data environments with their inconsistencies and governance constraints, is the central question the company has not yet answered in a commercial setting.

Why OpenAI's Backing Matters Beyond the Dollar Amount

OpenAI investing in Isara is not simply a financial transaction. It is a signal about where the leading AI lab sees the architecture of enterprise AI heading. OpenAI has its own agentic products under development, and its investment here suggests the company believes the multi-agent coordination problem is large enough that it will not be solved by any single organization. Backing external researchers working on complementary approaches is a way of maintaining visibility into an architecture that may prove important regardless of which company's models sit inside the agents.

For enterprise technology buyers, that signal matters. When the organization that arguably set the current trajectory of the AI industry places a significant bet on multi-agent coordination, it indicates that the single-model approach has real ceiling constraints that the industry is actively working to overcome.