Muse vs Dots: Consumer Trust vs Enterprise Plumbing

OpenAI's Dots and Meta's Muse landed within three weeks of each other in September 2026, and press coverage has understandably treated them as rivals answering the same question — should you let an AI agent keep a computer running on your behalf? But the sequencing of each launch reveals a real strategic disagreement about who answers that question first, and it shows up in almost every design decision that follows.

Who gets it, and in what order

Dots launched on September 29 to ChatGPT Pro and Business Premium subscribers, with the first dot free and enterprise, education, and healthcare deployments arriving through an admin-enabled beta. That's a top-down rollout: paying power users first, then IT-gated organizations, with the wider consumer base waiting. Muse took the opposite path. It launched September 8 on iOS, Android, and muse.ai, free for most features from day one, with paid tiers arriving later for heavier usage — and by late September had reportedly topped U.S. and Canadian app charts, outpacing what TechCrunch describes as ChatGPT's own early mobile trajectory. Meta went wide first and is now backfilling business use cases; OpenAI went deep into paying accounts first and will presumably go wide later.

Neither approach is obviously correct, but they imply different theories about trust. OpenAI's bet is that an agent this capable needs to prove itself with users who are already paying and already somewhat sophisticated about tooling — the Pro/Business Premium subscriber base — before it's handed to a general consumer who might not read a permissions screen carefully. Meta's bet is closer to its historical playbook with Instagram and WhatsApp: ship broadly, free, immediately, and let scale itself become the trust signal — if 100 million people are using it without incident, that's more persuasive than a careful staged rollout to enterprise IT departments.

Two different permission architectures

The products' safety models mirror this split. Dots relies on a layered set of controls sitting around the agent: Custom Rules that let a user require approval or prohibit specific actions outright, Auto-Review that checks consequential actions before execution, an explicit carve-out reserving sensitive operations like password changes for humans only, and an Activity View for after-the-fact oversight. It's a governance-first design — multiple named checkpoints a business admin or careful user can configure and inspect.

Muse takes a more architectural route: the agent runs inside an isolated Secure VM, and a separate agent called Sentinel approves or blocks every connected-app action and outbound network request at the system level, with Meta stating explicitly that "nothing Muse does reaches the internet unless the Sentinel approves it." There's no equivalent of Custom Rules — you're not writing your own policy — but there is a complete audit trail of what the agent has done and plans to do, and granular per-connection controls (read-only email access versus send access, for instance). We go deeper on what that audit trail actually shows in Audit Trails: How Much Visibility Does Muse Actually Give You?

The practical difference: Dots gives a sophisticated user or admin more levers to pull, which fits a rollout aimed first at Pro subscribers and business accounts who can be expected to configure them. Muse gives every user — including the ones who downloaded it from an app store on a Tuesday with zero context — a single hard boundary (Sentinel) they don't have to understand to benefit from. That's a defensible design for a free consumer product where most users will never open a settings menu, but it also means less individual control for the people who would want to fine-tune it.

Branding follows the same split

Both products lean into cute, humanized visual design — OpenAI gave Dots "little floating cartoons," a choice press coverage explicitly compared to Meta's approach, while Meta built a full Muse Realtime Avatar system for expressive video chat with a customizable voice. That convergence isn't a coincidence, and we cover why both companies landed on the same instinct in The Cartoon Problem, Again: Why Every Agent Gets a Friendly Face. But the underlying products diverge again on ecosystem: Dots plugs into 4,000+ existing applications through OpenAI's plugin infrastructure, while Meta says it received over 1,500 developer connector applications in the week after Muse's announcement — a nascent, Meta-controlled ecosystem still being built rather than an established one being extended.

What this means if you're choosing between them

If you're already paying for ChatGPT Pro or run a business on Microsoft's stack, Dots arrives with more configurable guardrails and a path into Agent 365 for enterprise identity and credential management — it's built for an environment where an IT department wants levers. If you're a consumer or a small business already living inside Instagram, Facebook, and Shopify, Muse arrives free, already wired into the tools you use, with a simpler trust model built around one hard boundary rather than many adjustable ones. Neither is more "agentic" than the other in any technical sense — they're the same underlying idea, aimed at two different first audiences, with permission systems built to match.

The unresolved question for both is the same one, though: an agent with its own persistent computer is only as trustworthy as the boundary around it, and in both cases, that boundary was described to us by the company selling the product. We don't yet have an independent account of either Sentinel or Auto-Review holding up under adversarial testing at scale.

Part 22 of 30 in the Meta Muse series. Previously: Ad Data and Agents: What Muse Promises to Keep Separate. Next: Muse vs Grok Bot: Two Different Bets on Who Adopts First. Browse more builder insights or explore AI skills at aiskill.market.