Durable Sessions AI UX vs System Design Architecture
// TL;DR
Use the Christensen Durable Sessions AI UX Framework if you are building or fixing a real-time AI chat or agent product and need resilient streaming, multi-device continuity, or live user control. Use the Simonyan System Design Architecture Skill if you need a general-purpose methodology for designing scalable backend systems from scratch — especially for interviews, architectural reviews, or greenfield services. Durable Sessions is a specialist framework for AI delivery infrastructure; System Design Architecture is a generalist framework for end-to-end system planning. Most teams need both at different stages.
// HOW DO THEY COMPARE?
| Dimension | Christensen Durable Sessions AI UX Framework | Simonyan System Design Architecture Skill |
|---|---|---|
| Best For | AI chat/agent products with streaming, real-time, and multi-device requirements | Designing any scalable backend system from scratch, especially for interviews or architectural reviews |
| Scope | Narrow: focuses exclusively on the delivery layer between AI agents and clients | Broad: covers APIs, databases, caching, load balancing, protocols, and full production infrastructure |
| Complexity | Moderate-to-high — requires understanding of pub/sub, WebSockets, and session-layer architecture | Moderate — builds progressively from a single-server baseline to distributed systems |
| Time to Apply | 1–3 days for an audit; 1–4 weeks for a full Durable Sessions implementation | 1–2 hours per design exercise; weeks to months for production implementation |
| Prerequisites | Existing AI/agent product with streaming; familiarity with SSE, WebSockets, and pub/sub patterns | Basic understanding of HTTP, databases, and client-server architecture |
| Output Type | A gap analysis of your AI UX delivery layer plus a redesigned session-based architecture | A complete system architecture diagram with articulated trade-offs for every major decision |
| Protocol Guidance | Deep: specifically diagnoses SSE limitations and prescribes WebSocket/bidirectional transport for live control | Broad: covers HTTP, WebSockets, gRPC, AMQP, TCP, and UDP selection criteria |
| Multi-Agent Support | Explicitly addresses orchestrator bottlenecks and multi-agent pub/sub patterns | Does not address AI agent architectures |
| Database Guidance | Not covered — assumes a persistence layer exists | Comprehensive: SQL vs NoSQL decision tree with sub-type selection (document, wide-column, key-value, graph) |
| Creator Background | Mike Christensen (Ably) — real-time infrastructure specialist, AI Engineer conference talk | Hayk Simonyan — system design educator focused on interview preparation and senior engineering practice |
What does the Christensen Durable Sessions AI UX Framework do?
The Christensen Durable Sessions AI UX Framework diagnoses why AI chat and agent experiences break under real-world conditions — network drops, multi-device usage, concurrent agents — and provides a prescriptive architecture to fix them. Its core insight is that most AI products rely on direct HTTP streaming (typically SSE), which couples the entire user experience to a single fragile connection. When that connection drops, the stream is gone.
The framework introduces the concept of a Durable Session: a persistent, shared resource sitting between agent and client layers. Agents write events to the session; clients subscribe to it. This single architectural inversion unlocks three foundational capabilities: Resilient Delivery (streams survive disconnections), Continuity Across Surfaces (sessions follow users across tabs and devices), and Live Control (users can steer or cancel agent work mid-generation).
The framework's 10-step workflow walks you from auditing your current streaming model through designing the Durable Sessions layer, replacing SSE with bidirectional transport where needed, flattening multi-agent architectures, and validating all three capabilities. It is specifically designed for teams that already have an AI product and need to harden its delivery infrastructure.
What does the Simonyan System Design Architecture Skill do?
The Simonyan System Design Architecture Skill provides a structured, progressive methodology for designing scalable backend systems from a blank slate. It starts with the simplest possible setup — one server, one database, one user — and systematically introduces complexity only when scale demands it.
The framework covers the full stack of backend architecture decisions: separating web and data tiers, choosing SQL vs. NoSQL databases (with a detailed sub-type decision tree), selecting scaling strategies (vertical vs. horizontal), configuring load balancers with the right algorithm, choosing API styles (REST, GraphQL, gRPC), selecting protocols (HTTP, WebSockets, AMQP), and eliminating single points of failure through redundancy and self-healing.
What distinguishes this skill is its emphasis on trade-off articulation. Every architectural choice must be accompanied by an explicit statement of what you gain and what you give up. This makes it especially valuable for system design interviews, where the ability to reason about trade-offs separates senior-level answers from junior ones. It is a generalist framework that applies to virtually any backend system.
How do they compare?
These two frameworks operate at different layers of the architecture stack and have almost zero overlap in what they cover.
Durable Sessions is a specialist framework. It assumes you already have a working AI product with an agent layer, a client layer, and a streaming mechanism. It does not help you choose a database, design a REST API, or configure load balancing. Its entire focus is the delivery and connectivity layer between AI agents and users — a layer that the Simonyan framework does not address at all.
System Design Architecture is a generalist framework. It helps you design the entire backend from scratch, including databases, APIs, caching, and infrastructure. It covers WebSockets and real-time patterns at a protocol-selection level but does not go deep into the specific challenges of AI agent streaming, session durability, or multi-agent pub/sub architectures.
On protocol guidance, Durable Sessions is clearly better if your specific problem is SSE limitations in AI streaming. It identifies the SSE Resume-Cancel Conflict — the fundamental ambiguity where closing an SSE connection could mean either "I disconnected, buffer my events" or "I pressed stop, cancel generation" — and prescribes bidirectional transport as the fix. The Simonyan framework covers WebSockets as one option among many but does not diagnose this specific failure mode.
On database selection, the Simonyan framework is clearly better. It provides a complete decision tree for SQL vs. NoSQL with four NoSQL sub-types. Durable Sessions has nothing to say about data persistence.
On multi-agent architecture, Durable Sessions is the only option. It directly addresses the Orchestrator Dual-Purpose Problem — where an orchestrator is forced to both coordinate sub-agents and relay their progress updates — and solves it by having every sub-agent write directly to the shared session. The Simonyan framework does not cover AI agent topologies.
On breadth of applicability, the Simonyan framework wins decisively. It applies to any backend system: e-commerce, social media, SaaS, fintech. Durable Sessions applies only to AI chat and agent products with real-time streaming requirements.
Which should you choose?
If you are building or auditing an AI chat or agent product and your users experience dropped streams, can't resume on another device, or can't interrupt a running agent — use the Christensen Durable Sessions AI UX Framework. It directly diagnoses and solves these problems with a specific architectural pattern. No other framework addresses this layer with this precision.
If you are designing a new backend system from scratch, preparing for a system design interview, or conducting an architectural review of a non-AI-specific system — use the Simonyan System Design Architecture Skill. It gives you a complete, progressive methodology from single-server baseline to production-grade distributed architecture.
If you are building an AI product that also needs a full backend architecture — use both. Start with the Simonyan framework to design your overall system (database, API, scaling, load balancing), then apply the Durable Sessions framework specifically to your agent-to-client delivery layer. They are complementary, not competing.
// FREQUENTLY ASKED QUESTIONS
Can I use the Durable Sessions framework for non-AI real-time applications?
The core pub/sub and session durability concepts apply to any real-time streaming application — collaborative editing, live dashboards, multiplayer experiences. However, the framework is specifically designed around AI agent patterns like multi-agent orchestration, LLM token streaming, and agent steering. For general real-time architecture, the Simonyan framework's WebSocket and protocol guidance is a better starting point.
Do I need to know system design basics before using the Durable Sessions framework?
Yes. The Durable Sessions framework assumes you already have a working AI product with an existing streaming architecture (SSE, WebSockets, etc.). It does not teach you how to set up servers, databases, or APIs. If you lack these fundamentals, start with the Simonyan System Design Architecture Skill first, then layer on Durable Sessions for your AI delivery layer.
Which framework should I use for a system design interview?
Use the Simonyan System Design Architecture Skill. It is explicitly designed for interview contexts with a progressive methodology that mirrors how interviewers expect candidates to think — start simple, scale incrementally, articulate trade-offs. The Durable Sessions framework is too narrow for general system design interviews unless the question specifically involves AI agent streaming.
Does the Simonyan framework cover AI-specific architecture patterns?
No. It covers general backend architecture — databases, APIs, load balancing, caching, protocols — but does not address AI-specific concerns like LLM token streaming, agent orchestration, multi-agent pub/sub, or the SSE Resume-Cancel Conflict. For AI delivery infrastructure, you need the Durable Sessions framework.
Can I replace SSE with WebSockets using either framework?
Both frameworks discuss WebSockets, but differently. The Durable Sessions framework gives you a specific diagnosis of why SSE fails for AI streaming (the Resume-Cancel Conflict) and prescribes WebSockets plus a Durable Sessions layer as the fix. The Simonyan framework covers WebSockets as a protocol choice for real-time features generally. For AI streaming specifically, the Durable Sessions framework's guidance is more actionable.
How long does it take to implement the Durable Sessions architecture?
The audit phase (steps 1–3) takes one to three days. Designing and implementing the Durable Sessions layer (steps 4–8) typically takes one to four weeks depending on your current architecture's complexity and how many of the three foundational capabilities you need. Validation and layering additional capabilities add another one to two weeks.
Which framework helps with multi-agent AI architectures?
Only the Durable Sessions framework addresses multi-agent architectures. It identifies the Orchestrator Dual-Purpose Problem — where the orchestrator is forced to both coordinate sub-agents and relay progress updates — and solves it by having each sub-agent publish directly to a shared Durable Session. The Simonyan framework does not cover agent topologies.
Are the Durable Sessions and System Design frameworks complementary or competing?
They are fully complementary. The Simonyan framework designs your overall backend architecture — database, API, scaling, load balancing. The Durable Sessions framework specifically designs the delivery layer between AI agents and clients. For an AI product, you would use the Simonyan framework for the backend foundation and then apply Durable Sessions to your agent streaming infrastructure.