Why Your AI UX Is Broken (and It's Not the Model's Fault)
summary
Argues that the real problem is connection architecture not the model and shows a concrete solution: decouple the session from any single connection so the session is a durable sessions shared resource. Uses Ably channels and an Ably AI Transport SDK built on pub/sub, replaces SSE vs WebSockets one-way limits, supports resumability, multi-tab sync, concurrent agents, a forced network disconnect that self-recovers, and mid-session handoff to a human with full history visible.
AI-written summary · about
chapters (13)
- 0:00Introduction to AI chat applications
- 0:51Current implementation: Direct HTTP streaming and SSE
- 3:03Three foundational capabilities for great AI products
- 4:34Limitations of direct HTTP streaming
- 5:21Introducing durable sessions
- 6:06Resumability in streams
- 7:43The conflict between SSE, resumability, and live control
- 9:13Multi-device and multi-tab synchronization issues
- 11:12Handling concurrent multi-agent architectures
- 12:54Using Pub/Sub and Ably channels for durable sessions
- 14:12Introducing Ably AI Transport SDK
- 15:34Live demo of durable session capabilities
- 17:38Handoff to human support agent
related talks
39mFrom Demo to Production: Why Agentic AI Systems Fail - and How to Fix Them
14mAI Sprawl: The Hidden Risk Threatening Your Enterprise Now
21mOvercoming Data Barriers for Successful AI Projects
34mAccountability in the Age of AI Agents
31mPanel: The AI Architecture Debt—Refactoring Infrastructure for Sustainable Privacy
22mYOU CAN'T AI YOUR WAY OUT OF A BROKEN PLATFORM
36mThe Best AI Strategy? Don’t Build AI 🤯 Build AI-Ready Software
27mAI is Eating the World, But Without AX, It’s Just Code