人工知能は、単発質問に応答するチャットアシスタントから、専用のクラウドPC、本物のデジタルアイデンティティ、長期タスク実行能力を備えた**常駐型「デジタル同僚（Always-on Agents）」**へと進化を遂げています。
This cheat sheet provides a side-by-side comparison of the 4 leading Always-on Agent architectures (Meta Muse, Grok Bot, OpenAI dots, and Manus Cue) across capabilities, runtime sandboxes, and enterprise selection criteria.
🔗 Related: AI Agent Skills · AI Agent Harness · Core AI Algorithms · Claude Code CLI
1. 4大 Always-on エージェントの全体概要
| Product | Parent Organization | Primary Product Stance | Distinctive Killer Feature | Typical Production Workloads |
|---|---|---|---|---|
| Meta Muse | Meta | Personal Life Assistant | Persistent Secure VM + Isolated browser sandbox | Calendar scheduling, retail price tracking, form automation, personal customer service triage |
| Grok Bot | xAI | AI Team Colleague | Full cloud PC (Linux terminal + browser + virtual filesystem) | Sales outreach, marketing pipelines, code debugging and bug fixing, 24/7 background telemetry monitoring |
| OpenAI dots | OpenAI | Long-Horizon Knowledge Work | Frontier reasoning models + Specialized Subagent swarms | In-depth market/scientific research, multi-week cross-functional initiatives, complex business workflows |
| Manus Cue | Manus | Autonomous Digital Identity | Independent email + dedicated phone number + digital wallet + private VM | Inbound/outbound voice calls, queuing/reservations, small-ticket discretionary payments, multi-agent group chats |
2. 8大中核次元における横断比較マトリクス
| Evaluation Dimension | Meta Muse | Grok Bot | OpenAI dots | Manus Cue |
|---|---|---|---|---|
| 1. Product Positioning | Personal lifestyle and life assistant | Technical/operations teammate for engineering & revenue | Dedicated long-horizon knowledge worker | Autonomous individual with an independent digital identity |
| 2. Persistent Runtime | Hardened Secure VM + browser sandbox | Full headless Cloud PC (Linux terminal + browser + persistent disk) | Managed cloud execution workspace + browser | Dedicated sandboxed personal computer per agent |
| 3. Real-World Execution | Web browsing, calendar booking, form filling, shopping checkouts | Cross-application, cross-website multi-step workflows operating 24/7 | Recurring periodic complex tasks executed over multiple days/weeks | Direct voice calls, SMS, reservation queues, discretionary debit payments |
| 4. Identity & Account Auth | Bound to user's personal social accounts (Email, Calendar, Instagram) | Operates user's existing accounts across web and SaaS tools | Deep enterprise integrations with customizable permission scopes | Owns dedicated independent phone number, email inbox, and crypto/fiat wallet |
| 5. Multi-Agent Swarms | Custom tools and linear workflows | Shared team memory across collaborators | Recursively delegates subtasks to specialized Subagents | Multiple autonomous agents join shared chat channels to collaborate |
| 6. Tool & App Integration | Integrates standard desktop and mobile productivity tools | Bridges arbitrary Web apps, email inboxes, and Linux CLI tools | Deeply integrated third-party Connected Tools ecosystem | Discretionary control over software installed in its virtual machine |
| 7. Safety & Guardrails | Mandatory user approval before purchases or emails; complete audit trails | Interactive confirmation prompts before destructive filesystem/network actions | Granular enterprise policies governing autonomous boundaries | Hard discretionary spend caps per transaction and calendar month |
| 8. Sweet Spot | Personal routine automation, lifestyle habit tracking | Growth marketing, customer prospecting, CI/CD triage, infrastructure maintenance | Deep industry reports, complex enterprise strategy, multi-team program delivery | Real-world phone calls, physical services logistics, multi-agent households |
3. システム基盤アーキテクチャの比較
| Architectural Layer | Meta Muse | Grok Bot | OpenAI dots | Manus Cue |
|---|---|---|---|---|
| Infrastructure Base | Isolated Secure VM | Cloud Linux container / virtualized TTY | Managed enterprise cloud execution | Dedicated virtualization sandbox instance |
| User Interface | Web / Mobile App / Social Graph | Web Workspace / Terminal CLI interface | Knowledge Kanban / Task Workspace | Voice telephony / SMS / Messaging apps / Dedicated email |
| Team Touchpoints | In-app alerts / Email digests | Enterprise channels / Slack | Enterprise taskboards and workspace threads | Telegram / Direct phone dial / Multi-agent groups |
| Long-Term Memory | Personal habit profiles and life goals | Team operational patterns and codebase context | Persistent task dependency trees and reference stores | Cross-task behavioral preferences and payment records |
4. 業務シナリオ別選定マトリクス：どのエージェントを選ぶべきか？
| Production Workload Category | Recommended Agent | Core Selection Rationale | Required Tools & Capabilities |
|---|---|---|---|
| Personal Lifestyle & Concierge | Meta Muse | Deeply embedded into consumer social and calendar graphs; safe automated checkout flows | Browser sandbox, verified audit trail, personal auth links |
| Software Engineering & DevOps | Grok Bot | Native Linux cloud terminal capable of executing bash commands, running tests, and managing repos | Shell CLI, virtual filesystem, persistent 24/7 background daemon |
| Enterprise Research & Strategy | OpenAI dots | Leverages top-tier reasoning models with automated Subagent task delegation | Subagent task graph, Connected Tools, enterprise permission guardrails |
| Real-World Physical Errands | Manus Cue | Holds legal phone numbers and prepaid spending cards; handles phone and delivery logistics | Virtual SIM telephony, digital debit wallet, multi-agent chat |
アーキテクチャ進化の本質と示唆
[The Dimensional Leap in Autonomous AI]
Previous Competition: Pure Foundation Model Benchmarks (Prompt Q&A ➔ Context Size ➔ Single Tool Invocation)
Current Competition: The Full-Stack Autonomous Engine (Cloud PC + Real Digital Identity + Tooling + Permissions + 24/7 Persistence)