Intuit is preparing to show how it rebuilt its AI infrastructure to support increasingly complex tasks—a shift that required tearing apart its existing setup and starting nearly from scratch. The company overhauled its technical platform after recognizing that customer expectations had moved well beyond simple conversational interactions. Users now expect AI agents that can handle complicated operations on their own, and legacy IT systems simply weren’t built for that kind of work.
At VB Transform 2026, set for July 14 and 15, the company’s VP of AI Nhung Ho will walk through the technology decisions behind the rebuild. Her presentation focuses on the abstraction layer Intuit built behind its system of intelligence.
The rebuild was not incremental.
Intuit moved away from its multi-agent setup—which prioritized broad capabilities across large, general-purpose components—to a granular framework built around specific skills and tools. They decomposed massive agents into specialized pieces, separating what she calls “the brain from the hands.”
Related: Query logs help AI agents fix SQL errors
“We went from a multi-agent system where we had large agents that did a lot to fully incorporating workflows, skills and tools down to the base level,” Ho said. “We changed the orchestrator, we changed the planner, we changed the brain, and we also changed what everybody had to build across the whole company.”
Human experts now work directly alongside AI.
Effective deployments don’t remove employees from the loop entirely—they put them closer to the action.
Intuit can now decouple orchestration from specific model providers.
Related: Hypernetworks and AI Autonomy Explained
The flexibility lets the company use whichever solutions work best, whether from large external vendors or its own home-grown code. Ho will detail how the abstraction layer works in practice and what tradeoffs came with dismantling the old architecture.
Enterprise Teams Rethink Orchestration at Scale
The talk is one of several at the event focused on agentic orchestration. Other sessions include a presentation by Target.
What Intuit’s Rebuild Reveals About Early Approaches
The decision to dismantle rather than layer capabilities on top of existing architecture carries lessons for other teams. Early multi-agent approaches may not hold up under real-world demand, since infrastructure built for conversational AI often fails to scale to the complex, multi-step tasks customers now expect.
Leave a Reply