Durable Sessions vs Autonomous App Building: Which Framework?

What does the Christensen Durable Sessions AI UX Framework do?

The Christensen Durable Sessions Framework diagnoses and fixes the real-time delivery layer of AI chat and agent products. Created by Mike Christensen of Ably, it starts from a core observation: most AI products stream responses over a single HTTP connection (typically SSE), and when that connection drops, the response is gone. This is the "Single-Connection Trap."

The framework introduces three foundational capabilities every production AI product needs: Resilient Delivery (streams survive disconnections), Continuity Across Surfaces (sessions follow users across tabs and devices), and Live Control (users can steer or stop agents mid-generation). The solution is a Durable Sessions layer — a persistent, shared pub/sub channel that sits between agents and clients. Agents write events to the session; clients subscribe to the session. Neither holds a direct pipe to the other.

This architecture eliminates the SSE Resume-Cancel Conflict (where closing a connection is ambiguous between "reconnect me" and "stop generating"), solves the Orchestrator Dual-Purpose Problem in multi-agent setups, and decouples agent logic from connection management entirely. The 10-step workflow takes teams from auditing their current streaming model through designing, implementing, and validating a Durable Sessions architecture.

What does the Isenberg Autonomous App Building Framework do?

The Isenberg Autonomous App Building Framework is a step-by-step method for building persistent, agent-operated web apps and internal tools inside OpenAI Codex Sites. Created by Greg Isenberg, it targets solo founders, creators, and small teams who want apps that update themselves — where AI agents handle the ongoing data entry, scoring, and management without human editing.

The framework follows six steps: build the app shell via the @sites plugin, add persistent storage (Cloudflare D1) with a reviewed data model, create Safe Actions (named mutations like `add_idea` or `move_card` that constrain what agents can do), create a Codex Skill (a reusable instruction manual so any future chat knows how to operate the app), checkpoint with a Save Gate, and finally Prove the Loop by opening a brand-new chat, invoking the Skill, and confirming the live app updates autonomously.

The key innovation is the Safe Action boundary: agents can only call approved mutations, never arbitrary SQL. Combined with Skills as persistent context, this makes the app reliably operable by any future Codex chat without human intervention.

How do the Durable Sessions and Autonomous App Building frameworks compare?

These two frameworks operate at entirely different layers of the AI product stack and solve different problems.

Durable Sessions is an infrastructure architecture framework. It is relevant when you already have an AI product with streaming responses and need to make the delivery layer resilient, multi-surface, and controllable. It requires backend engineering expertise, pub/sub infrastructure, and potentially a transport protocol change from SSE to WebSockets. The payoff is a production-grade AI experience that survives real-world conditions — network drops, multi-device usage, concurrent multi-agent activity.

Autonomous App Building is a product creation framework. It is relevant when you want to go from zero to a working, agent-operated app in hours. It requires OpenAI Codex access and no traditional coding skills. The payoff is a functional app with persistent data and a proven autonomous operation loop — but it does not address real-time streaming architecture, multi-device session continuity, or advanced agent infrastructure.

The Durable Sessions framework is clearly better for teams with existing AI products that need to fix UX breakdowns caused by fragile streaming. The Autonomous App Building framework is clearly better for non-technical builders or small teams who want to ship an agent-operated tool quickly without infrastructure work.

There is no overlap in their core use cases. You would never use the Isenberg framework to fix a streaming architecture, and you would never use the Christensen framework to bootstrap a Codex Sites app.

Which should you choose?

Choose Durable Sessions if: You have an AI chat or agent product in production (or approaching production) and users experience lost responses on disconnect, cannot see live responses on a second device, or cannot stop/steer an agent mid-generation. You have engineering resources and need an architectural pattern to solve these problems systematically.

Choose Autonomous App Building if: You want to build a new app or internal tool from scratch — a CRM-lite, editorial calendar, idea board, or similar — and you want Codex agents to operate it for you going forward. You value speed-to-ship and autonomous operation over deep infrastructure engineering.

Choose both if: You are building a Codex-powered app (Isenberg framework) that will eventually need resilient, multi-surface real-time delivery (Christensen framework). In that case, start with Isenberg to ship the app, then layer Durable Sessions principles as you scale the real-time experience.

Can these frameworks be used together?

Yes, but sequentially rather than simultaneously. The Isenberg framework gets your app built and autonomously operated. As your product matures and you need real-time streaming that survives disconnections and works across devices, the Christensen framework provides the architectural blueprint for that next layer. They address different stages of product maturity and different layers of the stack.