The Drawbridges Go Up
The AI Era is Speedrunning the Web 2.0 Story: Integrations Will Be Tightly Governed
The chatter around MCPs has been reminding me of Web 2.0. MCPs have become a rallying point for a larger vision of the future, where LLMs are connected to all the data and apps they’ll ever need, allowing them to easily perform all the rote tasks we’d like to avoid.
Web 2.0 described a similar vision (among other visions), where interconnected services would allow apps to leverage all your data, no matter where it resides.
For a while, every app published an API. Consultants made bank helping governments and corporations set up their own APIs. Many API standards emerged and withered, and we waded through the various security issues until we stabilized on a set of norms. Things were pretty good!
But it didn’t last.
Once network effects crowded a few winners, the drawbridges slowly pulled up. Previously simple APIs evolved into complicated layers of access controls and pricing tiers. Winning platforms adjusted their APIs so you could support their platforms, but not build anything competitive. Perhaps the best example of this was Twitter’s 2012 policy adjustment which limited client 3rd party apps to a maximum of 100,000 users (they’ve since cut off all 3rd party clients).
The Web 2.0 dream of unfettered free exchange evolved into tightly governed one-way interfaces. Today there are plenty of APIs to support ad buying and posting on Facebook, Google, or Twitter, but barely any that allow you to consume data.
We’re already seeing a similar narrative play out with MCPs.
MCPs are a touchstone technology, a feature that embodies the dream for what could be. We want LLMs to be able to access our apps and data so they can effectively answer questions or perform tasks for us. The problem is each platform likely has its own AI, which they’d prefer we’d use.
Let’s look at what happened in the last two weeks:
- “Slack, a Salesforce-owned workplace messaging app, recently blocked other software firms from searching or storing Slack messages.”
- “X has changed its developer agreement to prevent third parties from using the platform’s content to train large language models.”
- Anthropic, “cut Windsurf’s direct access to Anthropic’s Claude AI models largely because of rumors and reports that OpenAI, its largest competitor, is acquiring the AI coding assistant.”
- “Google, the largest customer of Scale AI, plans to cut ties with Scale after news broke that rival Meta, opens new tab is taking a 49% stake in the AI data-labeling startup.”
The drawbridges are coming up quickly, accelerated by consolidation. Data remains a moat and our dreams of proliferating MCPs are likely a mirage (at least for non-paying users). MCPs, like the APIs of Web 2.0, will shake out as merely a protocol, not a movement. It will be a mechanism for easy LLM tool use, but those tools will be tightly governed and controlled.
Sure, build an MCP for your app or service. But don’t expect the platforms to let you compete easily.