"Meta wants to post for you after you die" was the coverage of US 12513102B2. Good for clicks. Misses the point. Meta was granted this patent in December 2025 for a system that trains a language model on a user's historical interactions (posts, comments, likes, DMs, voice messages) and deploys it to simulate that user's social media behavior autonomously. Filed in November 2023 by CTO Andrew Bosworth.

The press ran with the death angle because it's eerie and shareable. The patent text is more interesting. The primary use case is simulating users who are "absent from the social networking system," which includes breaks, inactivity, or death. The deceased framing reads like a broadening mechanism for the claims. What they actually built is a personalized LLM that maintains engagement on behalf of any user, for any reason.

So why does Meta care about automating individual user activity?

The acquisitions tell you. December 2025: Meta buys Manus for over $2 billion. General-purpose AI agent platform, hit $100M ARR eight months after launch. Meta said they'd integrate it into consumer and business products including Meta AI.

March 2026: Meta buys Moltbook, a social network for AI agents. Acqui-hire bringing Matt Schlicht and Ben Parr into Meta Superintelligence Labs. Schlicht and Parr previously co-founded Octane AI, a conversational commerce platform that automated personalized customer interactions for Shopify merchants via Messenger and SMS. They've been building AI-driven business communication tools since 2016.

I think these three things are connected in a way the coverage hasn't picked up on. The "digital ghost" framing is wrong. What Meta is building is infrastructure for AI agents that act on behalf of businesses across their platforms. The small business owner spending hours managing their Facebook and Instagram presence is the real target user. The e-commerce brand running customer conversations through WhatsApp is the real target user.

The Schlicht/Parr hire is what convinced me. Meta didn't buy Moltbook because AI agents chatting with each other is commercially interesting (Bosworth said as much in an Instagram Q&A). They bought the team that knows how to make AI agents useful for businesses. That's what Octane AI was.

I'll be honest about the limits here. Companies file patents for all kinds of reasons and most aren't strategic. Engineers get bonuses for filings. Legal teams build portfolios for cross-licensing leverage. Some patents exist because someone wrote a disclosure and nobody said no. Reading a single patent and concluding "this is their roadmap" is a mistake I've made before.

What I've found more useful is matching filings against spending. A patent by itself is noise. A patent plus a $2B acquisition plus an acqui-hire of people who built a related product for a decade starts to look like a pattern. Google Patents and USPTO are free. LLMs make the filings surprisingly accessible now.

Curious if anyone else reads patents this way, or if I'm connecting dots that aren't there. Especially interested to hear from anyone on Meta's business tools side.