Didi Reveals AI Xiao Di Multi-Agent Architecture at WAIC 2026: Ride-Hailing Satisfaction Up by Nearly 10 Points

At the WAIC 2026 AI plus Consumption Forum on July 18, Didi AI Application Lead Liu Jiaqi delivered a keynote on how AI is reshaping the ride-hailing experience, revealing significant operational metrics from the company AI agent deployment.

Liu highlighted that Didi users travel needs extend far beyond commuting, encompassing shopping, dining, entertainment, and tourism. Didi research shows that every 1 yuan spent on ride-hailing drives several times that in offline consumption, positioning transportation as a key connector to the broader consumer economy.

The centerpiece of the presentation was Didi AI agent Xiao Di, which now has over 90 scenario tags to understand and match diverse user needs. When a user says something like taking the elderly to the hospital, need easy entry and exit, the AI parses the implicit requirements for a roomier vehicle and smoother driving. This represents a paradigm shift from users adapting to product rules to products understanding user needs.

Technically, Xiao Di employs a decentralized multi-agent architecture that has improved core ride-hailing scenario satisfaction by nearly 10 percentage points. Didi built a personalized ride-matching system paired with a driver-side tag system that optimizes dispatch algorithms, matching passenger preferences while giving better-performing drivers more opportunities, creating a positive feedback loop.

Beyond personalized ride matching, Xiao Di has rolled out features including search along the route, multi-destination planning, nearby recommendations, and complex address pickup and drop-off suggestions. These allow users to discover services en route and plan multiple consumption touchpoints in a single trip.

Looking ahead, Liu suggested that transaction-oriented agents could become the new infrastructure for service consumption, with personalized needs satisfaction becoming a universal service. Didi also showcased its next-generation Robotaxi model R2, jointly developed with GAC Aion, at the WAIC exhibition. The company is working with industry partners and regulators to build trusted standards for AI-powered mobility services.