Why Anthropic, the agent company, still won't build a personal AI assistant

创业邦 · wechat · 2026-10-19

OpenAI's dots, Meta's Muse (2.8M installs in 12 days), Google's Gemini Spark, and xAI's GrokBot have made "always-on personal agents" the most crowded track of the season — yet Anthropic is absent. GeekPark analyzes why:

It's already there, just at work: Claude Tag (June) lives in Slack as a shared, memory-holding agent that proactively follows up; an internal version writes 65% of the product team's code. Plus desktop Cowork, Claude Code, and Conway — an agent environment found in 512K lines of accidentally leaked Claude Code source;
No consumer gateway: Muse lives inside WhatsApp, dots rides ChatGPT's 900M weekly users; Claude's consumer weakness means a personal agent would attack rivals' thickest moat;
The ledger decides: 80% of Anthropic's revenue comes from API/enterprise; annualized revenue hit $65B in July, passing OpenAI. Always-on agents are costly (OpenClaw's defaults burn 4.8M input tokens/day, $19/user/day at Claude Opus 5.5 pricing), so Anthropic prioritizes highest-paying enterprise users;
Safety and attachment concerns: Muse's internal test exposed private iCloud photos; dots' proactive research is read-only; Anthropic's 2025 study stressed Claude isn't designed for companionship (2.9% of 4.5M conversations were emotional); enterprises have guardrails, personal life doesn't;
The open question: personal-agent moats are accumulated context. Anthropic bets on work context, Meta/OpenAI on life context. BlackBerry-vs-iPhone history warns consumer gateways can strike back — and Conway traces in iOS Claude hint Anthropic may eventually head for the living room.

Original post →

Why Anthropic, the agent company, still won't build a personal AI assistant

More from AGI Musings