Intuit rebuilt its AI agent architecture twice in four months
Lessons from VB Transform 2026: Speed and customer-centric iteration
1Critical failure demanding rapid rebuild
Intuit's central orchestration layer caused context loss in natural language transfers between agents. Compounding errors in 10-agent chains crashed the system, forcing a complete architectural overhaul in just 60 days.
2Transformation born from customer pain
Specialist agent fleet created user confusion about which agent to choose. The automated routing solution lasted merely 3 months, pushing Intuit toward more radical customer-centric changes.
3Engineering buy-in challenges
The rebuild decision meant shelving hundreds of engineers' work. Nhung Ho's team convinced leaders with real-query demos while showing engineers the scalability benefits of shared skills over standalone agents.
4Innovative human-AI collaboration
The new system enables live handoffs to accountants mid-chat, preserving full context. Unlike generic 'consult a professional' disclaimers, Intuit's solution offers seamless transitions with specialized financial data permissions.
5Feedback revolution
Chat-based interactions increased customer feedback rates from 0.3% to near 100%, exposing teams to raw critiques like 'You suck.' Intuit developed special models to analyze this unprecedented feedback volume for product improvements.
6The power of rapid iteration
Intuit's experience proves the value of fast experimentation over perfection in AI development. Two major rebuilds in four months enabled agile responses to customer needs, setting an industry benchmark for adaptive architectures.
Original article: VentureBeat — Intuit, AI ajan mimarisini 4 ayda 2 kez yeniden inşa etti