Mike Christensen leads product engineering at Ably, building Ably AI Transport, Pub/Sub, and LiveObjects. His work addresses a foundational problem for interactive AI: keeping agents, users, and devices synchronized when connections fail, conversations move between screens, or humans need to intervene.
Earlier in his career, Christensen worked at Discovery Education before becoming a technical lead at Blokur, where he tackled complex music-data systems. He studied at the University of Bristol and subsequently brought his experience with demanding data models and product engineering to Ably’s realtime infrastructure.
Christensen argues that unreliable AI experiences often stem from transport architecture, not deficient models. Streaming directly to a browser through HTTP or server-sent events couples generation to one connection, leaving applications responsible for buffering, sequencing, replay, and reconnection. His alternative replaces that fragile link with persistent, independently addressable sessions that agents and clients access separately.
- Durable sessions: Agents publish events into a shared session while clients reconnect and recover missed activity independently, avoiding bespoke recovery logic for every connection.
- Bidirectional agent control: Because closing a one-way stream can signify either cancellation or disconnection, explicit session events let users stop, redirect, or continue an agent without sacrificing resumability.
- Multi-device session continuity: Shared sessions synchronize browser tabs, phones, concurrent agents, and human support representatives. His AI Engineer Europe demonstration shows specialized agents reporting progress independently and a human joining with the complete conversation history.
- Long-running agent runs: AI Transport v0.2.0 models agent work as events spanning multiple steps, lifecycle transitions, interruptions, and alternative conversation branches.
- Database hydration: AI Transport v0.4.0 combines completed runs stored in application databases with newer session activity, giving agents and clients a consistent history while preserving application-owned persistence and analytics.
