Dangerous interoperability
This morning The Information ran a story about SAP’s “move to block OpenClaw”. The headline is misleading because SAP is really just trying to manage their computer resources against increased demand, but the concept is front of mind: companies are all trying to figure out how they need to adapt to a world in which every customer is also a developer.
APIs became commonplace because allowing developers to extend your product suite with small customizations meant creating customer lock-in. But now that AI allows anyone to use the API, they actually do the opposite: an API removes barriers between different software service providers.
Right now we are living in a golden age of interoperability. All of my data is available to me in ways that were previously unthinkable:
All my playlist and play history can be extracted to a CSV via the Spotify API. I could move my playlists from Spotify to TIDAL and cancel my subscription.
My task management history can be aggregated from Todoist, Monday.com, and Trello, and combined. I can switch from one platform to the other or I can build a desktop app to view my history and manage my tasks going forward in a combined dashboard.
I can download all my health data from Apple Health and run interesting statistical analysis on it by building a custom iPhone app that exports the data to me. I can import it to Google and switch to Android.
Cloud software providers are stuck in a tough spot: providing access to my data is fundamental to their service. But if they provide access to my data, I can easily move to another service. If I don’t like their interface, I can build my own interface or move to another service.
The model providers need to solve this problem, too. Today I can easily move between Claude and Gemini and GPT-5.5 and even Gemma4 running on my own machine, taking with me my skills and MCP connectors.
Anthropic is flexing its muscles because their models are the best right now, and trying to encourage people to use Claude’s desktop app and custom harnesses like Claude Code through pricing incentives: if you use OpenClaw, you can’t use the subscription pricing model, you have to use the more expensive API.
Eventually someone is going to figure out how to create meaningful lock-in around software services. I don’t know what that answer will be, but it has to happen in order to create healthy businesses.
In the meantime, it’s a golden age for getting your data and doing interesting things with it.

Clawed code, for the fail?