In a candid discussion on a recent podcast, two Intuit engineers revealed that the single most difficult technical challenge facing modern engineering teams is orchestrating multiple AI agents to work together seamlessly within complex systems.
Chase Roossin, group engineering manager at Intuit, and Steven Kulesza, staff software engineer, detailed the mounting complexity of multi-agent AI coordination, calling it "the hardest problem in engineering right now." The conversation highlights a crucial bottleneck as companies race to deploy AI at scale.
Background
As organizations increasingly rely on autonomous AI agents to automate workflows, the need for these agents to collaborate without conflict has become urgent. Roossin explained that the challenge escalates dramatically when multiple agents must share data, negotiate tasks, and maintain consistency across interdependent processes.
"Each agent operates with its own objectives and context," Roossin said. "Getting them to align without introducing errors or deadlocks is a fundamental problem." Kulesza added that traditional coordination techniques like locking or central schedulers often break down under the scale required for real-world enterprise applications.
Intuit, the company behind TurboTax and QuickBooks, has been on the forefront of integrating AI into financial software. Their engineers have firsthand experience with the pitfalls of multi-agent systems.
Expert Insights on the Coordination Crisis
"It's not just about making one agent smart," Kulesza emphasized. "It's about having a system where agents can communicate, learn together, and dynamically adjust their behavior without a human in the loop."
Roossin elaborated: "We've seen agents lock up or produce conflicting outputs because they're optimized individually. The real breakthrough comes from designing protocols for inter-agent communication and conflict resolution."
The engineers pointed to emerging solutions like shared state models, negotiation algorithms, and federated learning frameworks. However, they cautioned that no off-the-shelf tool currently solves the problem completely.
What This Means
The implications extend beyond Intuit. Any company deploying multiple AI agents—from customer service bots to supply chain optimizers—will face the same coordination hurdles. Without robust multi-agent orchestration, enterprises risk inefficiency, data inconsistency, and even cascading failures.
Industry experts predict that mastering multi-agent coordination could become the defining competitive advantage in AI adoption over the next two years. Startups and research labs are racing to develop new architectures and communication protocols tailored for agent swarms.
"We're at an inflection point," Roossin concluded. "The teams that crack this problem will unlock capabilities that are simply out of reach for the rest."
Future Outlook
Intuit is exploring modular agent frameworks with built-in negotiation layers. Kulesza noted that their internal projects now include "agent-to-agent" contracts—formal agreements that define how different AI agents interact. Early results show reduced conflict and improved system throughput.
For engineers and architects, the message is clear: invest in multi-agent coordination strategies now, or risk falling behind as AI systems scale. The conversation from Roossin and Kulesza serves as both a warning and a roadmap.
This breaking report is based on statements by Chase Roossin and Steven Kulesza from Intuit, made during a podcast episode. Additional context from industry sources is included.