Intuit Scrapped Its AI Agent Architecture Twice in Four Months
Summary
- Intuit was an early pioneer in the usage of agentic AI, but its path to success has hardly been a straight line.
- The company rebuilt its agent architecture twice in the span of about four months, first moving from a specialist agent system to a central orchestration layer, then abandoning that layer for a skills and tools based system.
- The failure mode that forced the second rewrite was specific, as agents in the orchestrated system passed results to each other in natural language, and each handoff lost context the next agent needed to act correctly.
- Intuit's AI VP, Nhung Ho, described how the company's original push toward specialist agents came from a straightforward customer complaint.
- A fleet of capable agents is still something a customer has to manage, deciding which agent to use for which task.
- Intuit's answer was a system that could take a task and route it internally, without asking the customer to pick an agent themselves.
- The clearest customer-facing result of the rebuild is a feature that lets a live agent conversation pull in a human, though it's currently in early testing, live to about 1% of Intuit's customer base.
Why It Matters
- The story highlights the challenges of building complex AI systems, where the architecture can break down due to structural reasons rather than capacity ones.
- It shows how companies are experimenting with different approaches to AI, such as skills and tools based systems, to improve efficiency and scalability.
- It also underscores the importance of customer feedback and testing in the AI development process.
- The story also matters because it shows how companies are trying to balance AI automation with human intervention, recognizing that sometimes, human judgment is needed to resolve complex issues.
- This trend is likely to continue as AI becomes more pervasive in various industries.
- Furthermore, the story highlights the importance of leadership buy-in and engineering buy-in in AI development, where convincing both leadership and engineers to adopt new approaches can be a significant challenge.
GenAI EXPLAINED
Agentic AI refers to AI systems that can perform tasks on behalf of a user, much like a human assistant. These systems are designed to be flexible and adaptable, allowing them to handle a wide range of tasks and situations. However, building agentic AI systems can be complex and requires careful design and testing to ensure that they work correctly.
A central orchestration layer is a system that coordinates the actions of multiple AI agents, ensuring that they work together seamlessly to achieve a common goal. However, this approach can break down if the agents are not designed to communicate effectively, leading to errors and inconsistencies.
Skills and tools based systems, on the other hand, focus on building individual skills and tools that can be combined to achieve complex tasks. This approach can be more scalable and efficient than traditional agent-based systems, as it allows companies to reuse and share skills and tools across different products and services.
MORE FROM THIS EDITION