Intuit Engineers Unveil Multi-Agent AI Coordination as Engineering's 'Hardest Problem'

Breaking: Multi-Agent AI Coordination Emerges as Top Engineering Challenge

Intuit engineers have identified what they call the single most difficult problem in modern engineering: orchestrating multiple AI agents to collaborate effectively within complex, large-scale systems. The revelation comes from a recent podcast interview with two senior technical leaders at the financial software giant.

Chase Roossin, group engineering manager, and Steven Kulesza, staff software engineer, detailed the mounting difficulties as companies deploy increasingly autonomous AI agents across interconnected platforms. The challenge, they warned, threatens to bottleneck the next wave of AI-driven innovation.

“The hardest problem in engineering right now is getting multiple agents to play nice at scale. It’s not just about building one smart agent—it’s about how hundreds of them coordinate without stepping on each other.”

Roossin and Kulesza underscored that the issue goes beyond simple technical integration; it involves complex trade-offs in resource allocation, decision sequencing, and conflict resolution. Without a robust coordination framework, agents can produce contradictory outputs or cause system-wide failures.

Steven Kulesza elaborated on the real-world stakes: “In a financial system like ours, even a minor misalignment between agents could cascade into a major error. We’re talking about millions of users affected in seconds.”

Background: The Rise of Agentic AI

The problem emerges against a backdrop of rapid adoption of AI agents —programs that can independently perceive their environment, make decisions, and take actions. Unlike traditional single-agent models, multi-agent systems are now being deployed in fields from finance to logistics, where autonomous entities must cooperate or compete.

Current approaches, such as using a centralized controller or enforcing rigid communication protocols, have proven inadequate at scale. Agents may struggle with shared memory, conflicting goals, or unpredictable interactions. The Intuit engineers argue that this is a fundamental research problem that requires new paradigms.

What This Means for the Tech Industry

If left unsolved, the coordination gap could slow the rollout of advanced AI systems that promise to revolutionize industries. For enterprises relying on multiple AI agents—such as customer service bots, fraud detection modules, and recommendation engines—the risk is that gains from individual AI capabilities are offset by integration failures.

Roossin offered a glimpse of a potential solution: “We need to think about hiring a ‘conductor’ that doesn’t just manage agents but learns from their interactions. This is more like developing a new operating system for AI.”

The engineers called for collaborative industry research and open standards to address the problem. Kulesza noted: “No single company can solve this alone. It’s a systems-level challenge that demands shared insights.”

Learn more about the background of multi-agent AI and understand the industry implications.