This summary argues that the real story behind Anthropic's leak was not the source code itself, but what it revealed about Conway: an always-on agent environment that hints at a much broader platform strategy. If that strategy succeeds, competition will move from chatbots to persistent agent layers that learn how people work and make switching costs much higher.

1. The Overlooked Conway Leak

The author says the leak mattered because it exposed Conway, an internal always-on agent project that had not appeared on any public roadmap. That matters more than the raw code because it points to Anthropic's strategic direction.

2. What Conway Looks Like

Conway is described as a dedicated agent environment inside Claude rather than a simple chat UI. It appears to have search, chat, and system areas, plus an extension model that feels a lot like an app store for agent capabilities.

3. A Tuesday Morning With Conway

The imagined use case is powerful: Conway triages email, drafts low-risk replies, watches Slack, surfaces relevant documents, and prepares meetings before the user types anything. Even if some of its output is wrong, the speed advantage could still make it valuable.

4. Anthropic's 90-Day Platform Push

The author frames Conway as one part of a coordinated platform move, alongside Claude Code, enterprise collaboration layers, distribution, and stronger control over where third-party tools can run. The point is not a collection of features, but a stack.

5. The Microsoft Parallel

This strategy is compared to Microsoft's climb from operating systems to office software to enterprise lock-in. Conway plays the role of the sticky internal layer that makes the rest of the stack harder to replace.

6. Open MCP, Closed Layers

Anthropic can support open standards like MCP while still building proprietary value on top. The comparison is Android versus Google Play Services: open foundations, but the most valuable distribution and workflow layers remain closed.

7. The App Store Choice for Developers

Developers may face a familiar decision: build portable MCP tools that work everywhere but lack built-in distribution, or build Conway-specific extensions that are less portable but easier to discover. History suggests app-store gravity often wins.

8. Why Behavioral Lock-In Is Different

Older lock-in was mostly about data. Conway-style lock-in would be about a model of how you work: what you ignore, what you prioritize, how you answer, and what your routines look like.

9. Intelligence Portability Is Still Unsolved

That work-pattern knowledge cannot be exported as easily as files or CRM records. If you leave the platform, you may lose months of accumulated behavioral context, not just software access.

10. The Coming Battle for the Always-On Layer

The next major AI platform fight may be about who owns the persistent layer that watches, coordinates, and acts across a user's workflow. Whoever controls that layer can shape the surrounding ecosystem.

11. Choosing an Agent Platform Carefully

This means companies should not treat all agent products as interchangeable. The choice of platform may determine where internal tools, extensions, and organizational habits ultimately accumulate.

12. Career and Company Implications

For individuals and teams, the practical question is no longer just which model is smartest today. It is which ecosystem is most likely to become the place where work habits, tools, and leverage compound over time.

Closing

The essay's central warning is simple: persistent agents may become the next major lock-in layer. Once they start learning how we actually work, switching costs could become much deeper than moving documents or APIs.