The VC Who Backed Both OpenAI and Anthropic Just Called xAI a 'Complete Disaster'
Reid Hoffman doesn't usually say the quiet parts out loud. So when the LinkedIn founder — and early investor in both OpenAI and Anthropic — calls xAI "a complete disaster" on a podcast, it's worth slowing down to actually listen.
He had a lot to say. And not all of it is the kind of thing you'd expect from someone whose portfolio is in the middle of all this.
What He Actually Said About xAI
According to Hoffman, xAI — Elon Musk's AI company behind the Grok model — has now lost all 11 of its original co-founders. Not a few departures. The entire founding team, gone.
And on benchmarks, Grok is falling behind both Anthropic and OpenAI. Not dramatically, not in a headline-grabbing way — just quietly, consistently behind. Given the resources Musk has poured in, and the hype around Grok's release cycles, that's a striking verdict from someone who's watched a lot of AI companies rise and struggle up close.
The SpaceX-Cursor Deal Doesn't Impress Him
SpaceX recently acquired Cursor, the AI coding tool that's become a go-to for developers. Hoffman's read: SpaceX is "not an AI company," and buying Cursor was essentially buying relevance — a large tech company trying to look AI-native by acquiring something that already is.
He also suggested Cursor may be past its peak. That's a bold call about one of the fastest-growing developer tools in years. But his argument tracks: if AI coding becomes a commodity feature baked into every IDE and every platform, the standalone-tool advantage disappears fast. The market isn't rewarding what Cursor was two years ago; it's rewarding what comes next.
His Defense of Anthropic — and His Frustration With the Government
Here's where it gets more layered. On June 11, the US government forced Anthropic to take down its Fable and Mythos models, citing export control concerns after Amazon's CEO reportedly flagged a jailbreak vulnerability in Fable 5.
Hoffman called the move "arbitrary and capricious." For an investor, that's unusually direct. His critique is about process: pulling advanced AI models based on a single flagged vulnerability, without apparent policy framework or due diligence, isn't a measured response. It's a panic response dressed up as regulation.
He still sees enormous growth ahead for both Anthropic and OpenAI. In his view, they're the two real contenders — actually building toward AI that works, not just racing for press cycles.
The Jobs Question
Hoffman didn't just talk company drama. He also pushed back on the AI-kills-jobs narrative directly: his advice to young people was not to resist AI.
That holds up against recent data. A firm called SignalFire tracked 80 million companies and found engineering is actually the most resilient job category of 2025. Total hiring at major tech companies is down 25% from 2019 levels, but engineering jobs have only dropped 11%. At early-stage startups, engineering hiring is up 7% compared to 2019.
NVIDIA's Jensen Huang put it differently: AI makes engineers busier, not redundant. The demand for people who can actually direct and build with AI is going up, not down.
What This Means If You Use OpenClaw
Hoffman's underlying thesis is simple: in the AI race, there are two genuine contenders — Anthropic and OpenAI. Everyone else is buying relevance or hoping hype does the work.
OpenClaw runs on Claude, Anthropic's model. When Hoffman says Anthropic has enormous room to grow, he's also saying the foundation your agents are built on is improving fast — and deliberately.
The Cursor point is also worth sitting with. Cursor is a coding agent. OpenClaw is a different kind of agent — one built for persistent, scheduled, automated work that runs without you watching it. The coding-agent category is real and valuable, but it's one slice. The broader category of agents that run on your behalf, hold context across sessions, and actually complete multi-step work without hand-holding — that's where the momentum is going.
Hoffman sees xAI failing because it couldn't hold onto the people who built it. His implicit argument is that what makes AI companies work is human judgment — the team behind the models. That's the same bet OpenClaw makes: give people the right tools, let agents amplify what humans can do, and build something that actually keeps working when you close the laptop.