Durable Sessions vs Agent-First Business Builder: Which?

What does the Christensen Durable Sessions AI UX Framework do?

The Christensen Durable Sessions Framework diagnoses and fixes the most common infrastructure failures in AI chat and agent-driven product experiences. It starts from a core observation: most AI products use direct HTTP streaming (typically SSE via tools like the Vercel AI SDK), which couples the entire response stream to a single client connection. If that connection drops — a mobile network switch, a closed laptop, a browser refresh — the stream is gone.

The framework introduces the concept of Durable Sessions: a persistent, stateful, shared layer between agents and clients. Agents write events to the session; clients subscribe to it. Neither holds a direct pipe to the other. This single architectural change unlocks three foundational capabilities: Resilient Delivery (streams survive disconnections), Continuity Across Surfaces (sessions follow users across tabs and devices), and Live Control (clients can steer, interrupt, or cancel agents mid-generation).

It also addresses the SSE Resume-Cancel Conflict — the fact that closing an SSE connection is ambiguous between a network drop and a user-initiated cancel — and the Orchestrator Dual-Purpose Problem, where orchestrators in multi-agent systems are forced to relay sub-agent updates instead of focusing on coordination.

This framework is purely an infrastructure and architecture skill. It does not address what your agents do or how your business runs — it ensures that whatever your agents produce actually reaches users reliably.

What does the Howie Liu Agent-First Business Builder do?

The Howie Liu Agent-First Business Builder is a complete operational playbook for designing, deploying, and scaling a fleet of AI agents that runs real business functions. It targets solo operators and small teams who want to replace or augment human labor — content marketing, deal flow analysis, market research, customer support — with autonomous agents powered by frontier models.

The framework is built around several key primitives. Skills are reusable instruction sets that teach agents how to perform specific jobs. Rubrics define what great output looks like and trigger automated LLM-as-Judge scoring after every run. The Command Center is a fleet-level dashboard for managing all agents, their schedules, quality trends, and deployment targets. The Self-Improvement Loop ensures agents accumulate memories and suggest skill updates, which the human operator curates over time.

The workflow begins with opportunity validation (the agent acts in "Founder Mode" to research and validate before building), progresses through V1 artifact creation and iterative coaching, and scales to a fleet of role-partitioned agents — each with its own Skills, Rubrics, and run schedules. A key mental model is the Human Equivalent Time Cost Reframe: evaluate a $150 token run against what a human would charge for the same output, not against a $10/month SaaS subscription.

This framework is a business-operations skill. It does not address how agent output reaches the user's device or how streams survive disconnections.

How do they compare?

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

The Durable Sessions Framework is an infrastructure architecture skill for engineering teams. Its domain is the delivery layer: how do agent-generated events reliably, resumably, and controllably reach one or more client surfaces? It assumes you already have agents producing output and focuses on the plumbing between agents and users. Its audience is engineers and architects building real-time AI products.

The Agent-First Business Builder is a business operations skill for founders and operators. Its domain is the agent layer and the business layer: what agents do you build, how do you define quality, how do you scale from one agent to a fleet, and how do you run a business on agent labor? It assumes the delivery mechanism is handled (or is simple enough not to need a custom session layer) and focuses on agent design, coaching, and fleet management.

If you are struggling with dropped connections, multi-device sync, or SSE limitations, the Durable Sessions Framework is clearly the right choice — the Agent-First Business Builder does not address these problems at all. If you are struggling with what agents to build, how to ensure quality at scale, or how to structure an agent-first business, the Agent-First Business Builder is clearly the right choice — the Durable Sessions Framework has nothing to say about business operations or agent design.

There is one scenario where both apply: you are building an agent-first product (Business Builder territory) that delivers real-time streaming output to end users (Durable Sessions territory). In that case, use the Business Builder to design your agent fleet and operations, and the Durable Sessions Framework to architect your delivery infrastructure.

Which should you choose?

Choose the Christensen Durable Sessions Framework if you are an engineer or architect working on an AI product whose streaming UX breaks under real-world conditions — disconnections, multi-device usage, lack of a stop button, orchestrator bottlenecks. This is your framework. It gives you a concrete architectural pattern (pub/sub-backed Durable Sessions) and a step-by-step audit-and-redesign workflow.

Choose the Howie Liu Agent-First Business Builder if you are a founder, solo operator, or small-team leader who wants to build and run a business on AI agent labor. You need to go from idea to deployed, self-improving agent fleet. This framework gives you the full lifecycle: opportunity validation, Skill creation, Rubric-based quality management, fleet scaling, and cost optimization.

Choose both if you are building a user-facing AI product where agent fleet operations and real-time delivery reliability both matter. Apply them to their respective layers — they are complementary, not competing.