Dark Factory Method vs Durable Sessions: Which Should You Use?
// TL;DR
These two skills solve completely different problems, so the choice is clear: use the Dark Factory Method if you need to orchestrate multiple AI coding agents to ship software faster, and use Durable Sessions if your AI product's streaming UX breaks on disconnects, multi-device use, or mid-generation user control. They are complementary, not competing. A team building an AI-powered product could adopt both — Dark Factory for their development workflow and Durable Sessions for their product's delivery architecture.
// HOW DO THEY COMPARE?
| Dimension | Koc Dark Factory Agent Shipping Method | Christensen Durable Sessions AI UX Framework |
|---|---|---|
| Best for | Orchestrating multiple AI coding agents to ship code at extreme velocity | Fixing broken AI chat/streaming UX so it survives real-world conditions |
| Problem domain | Developer workflow and agent management | Product infrastructure and real-time delivery architecture |
| Complexity | High — requires managing 5-20+ concurrent agent sessions, triaging backlogs, maintaining .skills files | Medium — architectural redesign of streaming layer, but well-scoped to one system boundary |
| Time to apply | Immediate if you already use AI coding agents; days to set up swim lanes and .skills library | Days to weeks depending on existing streaming stack; requires infrastructure changes |
| Prerequisites | Active codebase, work backlog, multiple AI coding agent sessions (Codex, Claude, etc.), test harness | Existing AI product with streaming responses, knowledge of current transport layer (SSE, WebSocket) |
| Output type | Shipped code, merged PRs, improved .skills files — a production process | Redesigned streaming architecture with resilient, multi-surface, controllable sessions |
| Who it's for | Engineers/tech leads managing AI agents as a coding workforce | Product engineers and architects building AI-powered user-facing products |
| Key mental model | Engineer as factory manager overseeing a production line of agents | Agent-client decoupling via a persistent shared session layer |
| Creator background | Vincent Koc — OpenClaw open-source project lead, AI engineering practitioner | Mike Christensen — Ably, real-time infrastructure and streaming architecture expert |
| Can they be used together? | Yes — use Dark Factory to build the product | Yes — use Durable Sessions to architect the product's delivery layer |
What does the Koc Dark Factory Agent Shipping Method do?
The Dark Factory method, created by Vincent Koc of OpenClaw, reframes the software engineer's role from craftsman to factory manager. Instead of writing code by hand, you orchestrate multiple autonomous AI coding agents (Codex, Claude, etc.) running in parallel across isolated "swim lanes" — each focused on a distinct work category like CI, feature development, bug fixes, or new P0/P1 issues.
The method provides a structured 9-step workflow: triage your backlog into swim lanes, instantiate agent sessions, set autonomy levels per lane, load reusable `.skills` files, monitor for agent waffling, use your test harness as ground truth, gate merges on taste rather than throughput, run evaluation loops, and retrospect on your `.skills` library.
The core insight is that in 2025–2026, the bottleneck is no longer your hands — it is your judgment. Token efficiency, not token maxing, is the discipline. You need to know when an agent is genuinely working versus bullshitting, and you need the courage to nuke a waffling session rather than rescue it.
What does the Christensen Durable Sessions AI UX Framework do?
The Durable Sessions framework, from Mike Christensen at Ably, diagnoses why AI chat experiences break under real-world conditions and prescribes a specific architectural fix. The core problem: most AI products use direct HTTP streaming (typically SSE) where the response stream is coupled to a single client connection. If that connection drops, the stream is gone.
Christensen identifies three foundational capabilities that separate a fragile demo from a great AI product: Resilient Delivery (streams survive disconnects), Continuity Across Surfaces (sessions follow users across tabs and devices), and Live Control (users can steer, interrupt, or cancel agents mid-generation).
The solution is a Durable Sessions layer — a persistent, stateful, shared resource sitting between agents and clients. Agents write events to the session; clients subscribe to the session. Neither holds a direct pipe to the other. This single architectural inversion unlocks all three capabilities simultaneously and eliminates the SSE Resume-Cancel Conflict, the Single-Connection Trap, and the Orchestrator Dual-Purpose Problem.
How do they compare?
These skills operate at entirely different layers of the stack and solve fundamentally different problems:
Dark Factory is a development process framework. It answers: "How do I manage 10+ AI coding agents working on my codebase without collapsing into chaos?" It is about shipping velocity, quality gating, and the human judgment required to direct autonomous agents.
Durable Sessions is a product architecture framework. It answers: "Why does my AI product's chat experience break when users switch networks, open a second tab, or press the stop button?" It is about infrastructure resilience, real-time delivery, and user experience.
Dark Factory is clearly better if your problem is development velocity and agent orchestration. Durable Sessions is clearly better if your problem is a fragile streaming UX that fails under real-world conditions. There is no overlap in what they solve.
However, they are highly complementary. A team using Dark Factory to ship an AI product at extreme velocity would benefit from applying Durable Sessions to that product's streaming architecture. You could even run a Durable Sessions redesign as one of your Dark Factory swim lanes.
Which should you choose?
Choose the Dark Factory Method if:
- You are already using AI coding agents (Codex, Claude Code, etc.) and want to scale from one session to many
- Your bottleneck is coordinating parallel agent work, not writing code yourself
- You need a process for triaging, monitoring, and quality-gating agent output
- You manage an open-source project or team drowning in PRs and issues
Choose the Durable Sessions Framework if:
- You are building an AI-powered product with a chat or streaming interface
- Users report lost responses, broken stop buttons, or inability to see conversations across devices
- You are using SSE (e.g., Vercel AI SDK) and hitting the Resume-Cancel Conflict
- You have a multi-agent architecture where the orchestrator is bottlenecking progress updates
Choose both if you are an AI product team that both develops with AI agents and ships AI-powered user experiences. Use Dark Factory for your development workflow and Durable Sessions for your product architecture. They sit at different layers and reinforce each other.
Neither skill is a substitute for the other. Picking the wrong one means solving a problem you don't have while the actual problem persists.
// FREQUENTLY ASKED QUESTIONS
Can I use Dark Factory and Durable Sessions together?
Yes, and they complement each other well. Dark Factory governs how you orchestrate AI agents to build software. Durable Sessions governs how your AI product delivers responses to users. A team building an AI chat product could use Dark Factory for their dev workflow and Durable Sessions for their product's streaming architecture. You could even run the Durable Sessions redesign as a swim lane inside your Dark Factory process.
Is the Dark Factory method only for open-source projects?
No. It originated in the OpenClaw open-source project but the swim lane methodology, .skills files, and agent monitoring principles apply to any software project where you run multiple AI coding agents in parallel. Solo developers building SaaS products, small startup teams, and enterprise engineering groups can all apply it. The key prerequisite is that you are already using AI coding agents and want to scale them.
Do I need Durable Sessions if I'm using the Vercel AI SDK?
Likely yes, if your product needs resilient delivery, multi-device continuity, or live user control (stop/steer). The Vercel AI SDK uses SSE by default, which creates the Single-Connection Trap: a dropped connection kills the stream. It also creates the Resume-Cancel Conflict where closing the connection is ambiguous between disconnect and cancel. Durable Sessions solves both by decoupling agents from clients.
What are .skills files in the Dark Factory method?
Dot skills files are versioned instruction files — similar to dotfiles — that encode how AI agents should behave for recurring task types like writing documentation, running tests, or committing code. They act as opinionated defaults so you don't re-explain context each session. You deploy them into agent sessions, review their output in logs, and update them after each heavy work session to compound efficiency gains over time.
What is the SSE Resume-Cancel Conflict?
SSE is one-way: the server sends data to the client, but the client's only control mechanism is closing the connection. This creates an irresolvable ambiguity: a closed connection could mean the user pressed stop (cancel) or the network dropped (resume). Cancel and resume require opposite responses from the server. Durable Sessions solves this by using bidirectional transport with explicit cancel signals instead of relying on connection closure.
How many AI agent sessions do I need to run the Dark Factory method?
Koc recommends 5–20 active swim lanes depending on your available brain-space. Each swim lane gets one or more agent sessions. You don't need to start at 20 — a solo developer can begin with three lanes (features, tests, docs) and scale up. The key constraint is not compute cost but your ability to monitor sessions and detect when agents are waffling.
Does Durable Sessions require WebSockets?
Not entirely, but WebSockets or an equivalent bidirectional transport are required if you need Live Control — stop buttons, steering messages, or follow-up prompts mid-generation. Resilient Delivery and Cross-Surface Continuity can be achieved with a pub/sub session layer over various transports. However, the full benefit of Durable Sessions is realized with bidirectional transport that supports explicit cancel signals.
Which framework is harder to implement?
Dark Factory is operationally complex but requires no infrastructure changes — you're organizing your existing AI agent usage into a disciplined process. Durable Sessions requires actual architectural changes to your product's streaming layer, including introducing a session substrate, redirecting agent output, and potentially replacing SSE with WebSockets. Durable Sessions is the harder infrastructure lift; Dark Factory is the harder behavioral and management shift.