Durable Sessions AI UX vs Ng ML Orientation: Which to Use?
// TL;DR
These two frameworks solve completely different problems and almost never compete. Use the Ng ML Orientation Framework when you need to evaluate whether machine learning applies to a problem and justify ML adoption. Use the Christensen Durable Sessions AI UX Framework when you are already building an AI-powered product and need to fix broken streaming, disconnections, multi-device continuity, or live agent control. If you are an AI product engineer shipping a chat or agent experience, start with Durable Sessions. If you are new to ML or scoping a project, start with Ng's framework.
// HOW DO THEY COMPARE?
| Dimension | Christensen Durable Sessions AI UX Framework | Ng Machine Learning Orientation Framework |
|---|---|---|
| Best for | AI product engineers fixing broken streaming and chat UX | Beginners or decision-makers evaluating if ML fits a problem |
| Problem stage | You already have an AI product and need to harden its delivery layer | You are at the very start — deciding whether ML is the right approach |
| Complexity | High — requires understanding of SSE, WebSockets, pub/sub, agent architectures | Low — no technical prerequisites beyond basic problem description |
| Time to apply | Days to weeks for architectural redesign | 30 minutes to a few hours for problem framing |
| Prerequisites | Working AI/agent product, streaming infrastructure, engineering team | A real-world problem statement; no code or data required to start |
| Output type | Architectural redesign plan with a Durable Sessions layer | Problem classification and ML justification statement |
| Creator background | Mike Christensen (Ably) — real-time infrastructure for AI products | Andrew Ng (DeepLearning.AI, Stanford) — foundational ML education |
| Scope of impact | Transforms product reliability, multi-device UX, and agent control | Clarifies whether ML is worth pursuing for a given problem |
| Audience technical level | Senior engineers, architects, AI product leads | Anyone — non-technical stakeholders, students, product managers |
| Overlap | None — assumes ML/AI is already chosen and running | None — does not address product delivery or UX infrastructure |
What does the Christensen Durable Sessions AI UX Framework do?
The Christensen Durable Sessions AI UX Framework diagnoses and fixes the delivery layer of AI-powered products — specifically chat interfaces and agent-driven experiences. It identifies a core architectural problem Mike Christensen calls the Single-Connection Trap: when your AI product streams responses via direct HTTP (typically SSE), the stream's health is tied to a single client connection. If that connection drops, the response is lost.
The framework prescribes an architectural redesign around Durable Sessions — persistent, shared, stateful resources that sit between your agent layer and your client layer. Agents write events to the session; clients subscribe to it. This decoupling 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).
This is a deeply technical framework. It addresses SSE's fundamental inability to distinguish between a network drop and a user-initiated cancel, solves the orchestrator relay bottleneck in multi-agent systems, and provides a pub/sub-based architecture that scales. If you ship an AI product and users lose responses when switching networks, cannot see the same conversation on two devices, or your stop button is unreliable, this framework tells you exactly why and how to fix it.
What does the Ng Machine Learning Orientation Framework do?
The Ng ML Orientation Framework helps anyone — from students to executives — determine whether machine learning is the right approach for a given problem. Derived from Andrew Ng's foundational Machine Learning Specialization, it provides a structured six-step process for evaluating ML applicability.
The workflow is straightforward: state the problem in plain language, check if explicit programming could handle it, map the problem to known ML application patterns (classification, recommendation, anomaly detection, etc.), identify the learning signal (data), classify the problem on a consumer-to-industrial spectrum, and articulate why ML is justified.
This is a problem-framing tool, not a build tool. It does not teach you how to train models or ship products. Its value is in preventing two costly mistakes: applying ML where simple rules suffice, and assuming ML will work without verifying that training data actually exists. It is the right starting point for anyone who has not yet confirmed that ML is the correct solution.
How do they compare?
These frameworks operate at completely different stages of the AI product lifecycle and solve fundamentally different problems. They do not compete.
Ng's framework is upstream. It answers: Should we use ML at all? It requires no code, no infrastructure, and no technical background. It produces a decision and a justification.
Christensen's framework is downstream. It answers: Why does our AI product break under real-world conditions, and how do we fix the delivery architecture? It requires an existing AI product, an engineering team, and deep familiarity with streaming protocols.
On complexity, the Durable Sessions framework is significantly harder. It involves architectural patterns like pub/sub channels, WebSocket transports, agent-client decoupling, and multi-agent session flattening. Ng's framework is intentionally accessible and can be applied by a product manager in a meeting.
On time to apply, Ng's framework takes under an hour. Christensen's framework drives weeks of engineering work.
On output, Ng's framework produces a clear yes/no ML decision with justification. Christensen's framework produces an architectural migration plan that fundamentally changes how your product delivers AI-generated content to users.
The only scenario where both are relevant to the same person is a technical founder or product lead who is first evaluating whether to build an ML-powered product (Ng) and later needs to fix its UX infrastructure (Christensen). But they would apply them months apart.
Which should you choose?
Choose the Ng ML Orientation Framework if:
- You are deciding whether ML applies to your problem
- You need to justify ML adoption to stakeholders
- You are onboarding someone new to machine learning concepts
- You have a problem but no technical approach yet
Choose the Christensen Durable Sessions AI UX Framework if:
- You already have an AI chat or agent product in production or development
- Users experience lost responses, broken streams, or unreliable stop buttons
- You need multi-device session continuity
- You are building multi-agent systems and the orchestrator is becoming a bottleneck
- You are using SSE (e.g., Vercel AI SDK) and hitting its limitations
If you are building an AI product and the UX feels fragile — responses vanish on mobile, the stop button is unreliable, a second tab cannot see the live response — the Durable Sessions framework is what you need, and it is clearly the more impactful framework for AI product teams. The problems it solves are the ones that separate demos from production-quality AI products.
If you are not yet at the build stage, Ng's framework is the correct and only relevant starting point. It is the better framework for ML problem framing, and no amount of infrastructure design matters if you have not confirmed ML is the right approach.
Neither framework replaces the other. Use them sequentially if your journey takes you from problem identification all the way to production AI product delivery.
// FREQUENTLY ASKED QUESTIONS
Can I use both the Durable Sessions and Ng ML Orientation frameworks together?
Yes, but sequentially rather than simultaneously. Use Ng's framework early to validate that ML is the right approach for your problem. Months later, when you have a working AI product and need to fix its streaming UX and reliability, apply the Durable Sessions framework. They address completely different lifecycle stages.
Which framework should I use if I'm building an AI chatbot?
If your chatbot is already built and users experience dropped responses, broken stop buttons, or cannot continue conversations across devices, use the Christensen Durable Sessions framework. If you have not yet confirmed that an AI chatbot is the right solution for your problem, start with Ng's ML Orientation Framework.
Do I need to know how to code to use the Ng ML Orientation Framework?
No. The Ng ML Orientation Framework is explicitly designed for non-technical users as well as engineers. It requires only a plain-language problem description. You need no code, no data, and no ML experience to apply its six-step workflow and determine whether ML fits your use case.
What is a Durable Session in AI product architecture?
A Durable Session is a persistent, stateful, shared resource that sits between AI agents and client applications. Agents write events to it; clients subscribe to it. It survives connection drops, supports multiple devices viewing the same session, and enables live user control of agents — solving the core fragility of direct HTTP streaming.
Is the Durable Sessions framework only for products using SSE?
No, though SSE-based architectures benefit most because SSE has a fundamental resume-cancel conflict. The framework applies to any AI product using direct HTTP streaming — including raw WebSocket implementations that still couple stream health to a single connection. The core issue is the Single-Connection Trap, which affects any direct streaming model.
Which framework is better for a startup evaluating AI opportunities?
Start with the Ng ML Orientation Framework. It will help you determine whether ML applies to your target problem, what data you need, and how to justify the approach. The Durable Sessions framework becomes relevant only after you are actively building an AI product and need to ensure its real-time UX is production-grade.
Does the Ng ML Orientation Framework teach me how to build machine learning models?
No. It is a problem-framing tool that helps you decide whether ML is applicable and justify that decision. It does not cover model selection, training, or deployment. It comes from the first lesson of Andrew Ng's ML Specialization — the actual model-building content follows in subsequent lessons and courses.
What problems does the Durable Sessions framework solve that WebSockets alone cannot?
WebSockets provide bidirectional communication but do not inherently solve multi-device visibility, message persistence across disconnections, or session resumability. The Durable Sessions framework adds a shared, persistent session layer on top of any transport, so multiple clients see the same live activity and agents never manage connection state directly.