The Switchboard
AI news, translated for operators.
The Big Take
Zuckerberg Building a Personal AI Agent to Run Meta
Meta is developing internal AI tools across the company, but the most revealing project is personal: Mark Zuckerberg is building a "CEO agent" to help him get information faster and make decisions. The tool reportedly synthesizes data across Meta's sprawling operations on demand.
What this means for you: This is Jensen Huang's "harness" thesis in action at the highest level. If a CEO running a trillion-dollar company sees value in a personal AI agent, the pattern will cascade down. Start thinking about what an executive-level AI harness looks like for your organization. The OpenClaw-style setup is becoming standard operating procedure, not an experiment.
For Operations & RevOps Leaders
The Generalist Becomes the Human Trust Layer in the AI Era
AI is enabling employees to work beyond their core expertise. Anthropic found that 27% of AI-assisted work represents tasks that wouldn't have been attempted otherwise. But the hallucination problem creates a new role: the generalist as quality control.
The most effective AI users aren't specialists. They're people with broad fluency who can spot when AI output is confidently wrong. They become the human trust layer sitting between AI output and organizational standards, deciding what passes and what needs expert review.
What this means for you: This connects directly to why middle managers matter in AI implementation. The companies that succeed aren't eliminating generalists; they're repositioning them as AI oversight. When hiring, prioritize curiosity and critical thinking over narrow specialization. The ability to catch hallucinations is now a core competency.
The AI Stack
Why AI Progress Will Be "Lossy," Not Recursive
The recursive self-improvement narrative assumes AI can improve itself in a closed, self-amplifying loop that leads to exponential capability gains. Nathan Lambert at Interconnects argues the reality will be different: lossy self-improvement, where friction breaks down the core assumptions.
The complexity brake matters. Microsoft co-founder Paul Allen argued that the more progress science makes toward understanding intelligence, the harder additional progress becomes. Patent data shows human creativity peaked in 1850-1900 and has declined since. Automatable research remains too narrow. There's a gap between lower test loss and models users find more productive.
What this means for you: This is a useful counter to the breathless singularity coverage. AI will keep getting better, but the improvement curve is more likely sigmoid than exponential. Plan for sustained progress measured in years, not sudden discontinuities. The complexity of real-world deployment will be the limiting factor, not raw capability.
For Media & Publishing Leaders
Google Testing AI-Generated Headline Rewrites in Search
Google confirmed it's testing AI-generated headline rewrites in Search results, following the same playbook that made AI headlines a "feature" in Discover within a month. In one documented case, "I used the 'cheat on everything' AI tool and it didn't help me cheat on anything" appeared as "'Cheat on everything' AI tool." No disclosure is shown to users.
Google's existing title rewrites pull from on-page elements. These new rewrites generate text that doesn't exist anywhere in the article. That's a fundamentally different intervention.
What this means for you: If this graduates from test to feature, publishers lose headline control across both Discover and Search. There's no opt-out mechanism. Start monitoring your Search appearances manually. More importantly, focus on building direct audience relationships that don't depend on Google's representation of your content.
The Web Needs to Speak to AI Agents: Enter AAIO
SEO optimized for search crawlers. AEO optimized for featured snippets. GEO optimized for AI-generated answers. Now comes AAIO: Agentic AI Optimization, where your website needs to be usable by AI agents that act autonomously.
In December 2025, the Linux Foundation announced the Agentic AI Foundation with Amazon, Anthropic, Google, Microsoft, and OpenAI as platinum members. Three protocols were contributed: Model Context Protocol (MCP), Goose, and agents.md. The competing tech giants are building shared infrastructure instead of fighting standards wars.
The progression: SEO asks "How do I rank?" AEO asks "How do I get cited?" GEO asks "How do I get included?" AAIO asks "How do I enable agents to complete tasks on my site?"
What this means for you: This is a five-part primer series worth following. If your website can't be parsed by AI agents, you'll be invisible to the growing share of web traffic that's machine-mediated. Start with the basics: structured data, clear machine-readable content, and awareness of MCP. The December announcement signals this is infrastructure, not hype.