For months, the AI industry has obsessed over model capabilities—whose LLM is smarter, faster, cheaper. But Peter Steinberger, an Austrian developer, identified what the labs missed: the bottleneck wasn't intelligence. It was reach.
OpenClaw (originally Clawed Bot before Anthropic's trademark threat) is elegantly simple. It's a layer of open-source connectors that lets any AI model—GPT, Claude, Gemini, or open-source alternatives—access tools and APIs without waiting for their respective companies to bless integrations. A Claude instance trained to use WhatsApp. A GPT that autonomously discovers which OpenAI models it can invoke to solve problems. Uber Eats orders placed through Telegram. The infrastructure of agency, detached from the gatekeepers.
Anthropic's response was predictable: block the model, protect the moat. OpenAI's was smarter: hire Steinberger, keep the project alive as separate infrastructure, harvest the goodwill. The move cost them nothing—the code is already public—and gained them developer loyalty plus a foundational layer for their ambitions around OS-level AI and their rumored "super app" that will consolidate ChatGPT, Code Interpreter, and other tools into one interface.
The real tension emerges here: OpenAI is consolidating control (one app, one gate), while simultaneously betting on open-source architecture (OpenClaw) that erodes vendor lock-in. The contradiction dissolves when you see it as strategy. If users can script their own integrations via OpenClaw, they'll stay on the platform not because they have to, but because OpenAI's models and infrastructure work best. Control through quality, not walls.
Meta paid for Multibot (an agent social network) less for the product than for the team. Mark Zuckerberg is building personal agents for all 70,000 Meta employees—a quiet bet that the future of productivity is not better software, but custom AI co-workers that know your company's systems. Anthropic, meanwhile, pivoted slightly: they're sunsetting Sora (video generation) to redirect compute toward Claude and its upcoming desktop interface, mirroring Lovable's approach. The pattern is clear. The labs are moving from "we have better models" to "we are the platform your model lives in."