Why Your AI UX Is Broken (and It's Not the Model's Fault) — Mike Christensen, Ably

2026-05-17 · Source: AI Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, long

Summary

The default pattern for AI chat applications, direct HTTP streaming using Server-Sent Events (SSE), fundamentally limits the quality and richness of user experiences. This approach couples the response stream to a single client connection, making it fragile to disconnections, unable to provide continuity across multiple devices or tabs, and lacking bidirectional control for live interaction with agents. Ably, a real-time messaging platform, identifies three foundational capabilities for superior AI experiences: resilient delivery, continuity across surfaces, and live control. To overcome the limitations of direct HTTP streaming, many engineering teams are adopting "durable sessions," a shared, persistent, and stateful resource that decouples the agent and client layers. This pattern enables resumable streams, multi-client interaction, and concurrent activity in multi-agent architectures, simplifying complex plumbing and enhancing user experience. Ably AI Transport is introduced as an SDK built on Ably channels to facilitate this durable session pattern.

Key takeaway

For CTOs and VPs of Engineering building AI-powered products, relying solely on direct HTTP streaming for chat applications will constrain product quality and user experience. You should evaluate adopting a "durable sessions" architecture to enable resilient delivery, seamless continuity across devices, and live agent control. This shift will simplify complex engineering challenges, reduce development overhead for features like stream resumability and multi-agent coordination, and ultimately allow your teams to focus on building richer, more interactive AI experiences.

Key insights

Direct HTTP streaming limits AI chat experiences; durable sessions enable resilient, multi-device, and interactive agent communication.

Principles

• Decouple agent and client layers for robust AI experiences.
• Persistent, stateful sessions enhance stream resilience and continuity.
• Bidirectional communication is crucial for live agent control.

Method

Implement durable sessions as a shared, persistent, and stateful medium between agent and client layers, leveraging pub/sub channels to manage event streams, enable resumability, and support multi-client/multi-agent interactions.

In practice

• Use WebSockets over SSE for bidirectional agent control.
• Store events with sequence numbers for stream resumability.
• Employ a pub/sub model for multi-client session synchronization.

Topics

• AI User Experience Design
• HTTP Streaming Limitations
• Durable Sessions
• Real-time Messaging Platforms
• Ably AI Transport

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, MLOps Engineer, AI Architect

Related on AIssential

• AI Agents Need Real-Time Context: Data Streaming Is How You Are Going To Get It — AI agents need real-time, pristine data context, best delivered by a unified streaming data platform.
• When a chatbot runs your store — Large language models struggle with reality adherence, making them unreliable for autonomous transactional roles.
• The TCP/IP Moment for AI Agents Is Here — AI agents require a dedicated session layer to overcome current web infrastructure limitations in connectivity, identity, and cost.
• How to Build a Memory Your AI Agents Can Actually Reuse — AI agents need reusable memory across sessions, not just more context, to avoid repetitive inputs.
• AI for Frontend Developers — Day 69 — AI applications evolve into information management systems, making robust conversation organization essential for usability and scalability.
• Claude Agent SDK Streaming: Your AI Agent Already Knows What It Is Doing. — A silent AI agent pause is a UX failure; streaming output provides live narration, improving user experience.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Engineer.