Personal AI agents are evolving from answering questions in chat to autonomously managing email, calendars, documents, and business workflows. OpenAI introduced Dots on September 29, and Anthropic opened research access to Claude Duet on October 21, 2026. Both products are capable of advancing tasks without an open chat window, utilizing connected applications, and requesting human confirmation for high-stakes actions. However, their focuses differ: Duet aims to be an attentive assistant throughout the current workday, while Dots is oriented toward long-term projects and persistent work between conversations.
On November 13, The Verge published a week-long hands-on review of Duet. The review highlighted the advantages of having such an assistant while simultaneously exposing a challenge that marketing presentations often gloss over: an agent can see a deadline, yet completely misjudge its true importance. To compare the two, we draw on this practical experience alongside official documentation from both companies. A direct, identical test of Duet and Dots on live personal accounts is not conducted in this article.
What Claude Duet Can Do
Anthropic designed Duet as a dedicated proactive agent powered by Claude Opus 5.5. It accesses only user-authorized applications, monitors work events, compiles morning briefings, and suggests next steps. Users can assign it to organize meetings, draft replies, assemble documents, presentations, or dashboards. The work can continue on a cloud machine even after closing the app.
Unlike standard queries to Claude, Duet proactively surfaces reminders and suggests actions. Anthropic emphasizes the necessity of explicit confirmations before sending messages, making payments, signing documents, or deleting data. At the same time, the company cautions that the agent can misinterpret correspondence or misjudge urgency.
As of this writing, Duet is available in the US by invitation from a waitlist to Claude Pro, Max, and Ultra subscribers. This is a research preview rather than a universally available feature. Duet's usage limits are shared with the rest of Claude, including the consumption from background checks. The agent does not allow users to observe its internal browser directly or take manual control of it, and support for several integrations remains limited.
How ChatGPT Dots Works
Dots is OpenAI's agentic system introduced on September 29. A dot operates on a separate cloud computer, utilizing connected apps and memory from prior interactions. The user specifies goals and boundaries of autonomy, allowing the agent to research questions, generate documents, build applications, and maintain progress between sessions.
According to OpenAI's official documentation, Dots initially ran on GPT-6 Astra, with GPT-6.1 Terra added on November 3: the system can invoke different models for different phases of work. Thousands of integrations are accessible through the plugin ecosystem. A dot can be utilized within ChatGPT as well as supported workplace channels like Slack and Teams.
An important limitation: proactive background checks in Dots operate in read-only mode. They cannot unilaterally send messages, modify records in third-party applications, or steer the browser. Any active operations require explicit user instructions, permissions, and validation steps. This serves as an important clarification to the marketing impression of an agent that handles everything independently.
Access is similarly restricted. As of November 15, personal dots are gradually rolling out to adult ChatGPT Pro subscribers in supported regions, excluding the EEA, Switzerland, and the UK, with separate business and enterprise tiers available. Direct conversations with a dot do not consume regular chat limits, but tasks delegated to ChatGPT Work or Codex are metered separately. Commercial terms for enterprise access differ.
Claude Duet vs ChatGPT Dots: Comparison
| Criteria | Claude Duet | ChatGPT Dots |
|---|---|---|
| Core Concept | Proactive workday assistant | Long-term project and task agent |
| Technical Foundation | Claude Opus 5.5 | GPT-6 Astra, with GPT-6.1 Terra added |
| Memory and Context | Claude memory, history, and connected apps | ChatGPT context, dot memory, and plugins |
| Typical Scenarios | Morning briefings, messaging, meetings, documents | Research, development, content, long-running processes |
| Autonomous Work | Background checks and action execution with confirmations | Background research; actions require permissions and checks |
| Access as of 15.11.2026 | US, Pro/Max/Ultra, waitlist | Phased rollout on select tiers in supported regions |
| Usage Limits | Shared with Claude, including background checks | Chat with dot is separate from chat limits; Work/Codex delegation billed separately |
This comparison reflects announced capabilities and constraints rather than direct benchmarks. The raw number of integrations does not indicate how reliably an agent completes tasks within any specific application.
Real-World Experience: When Duet Helps and When It Gets in the Way
The Verge's Allison Johnson granted Duet access to email, calendar, Google Docs, and Slack. The agent successfully connected insights across different sources: for example, matching notes from a meeting with email threads to identify scheduling conflicts. In one instance, it recognized that upcoming vacation days reduced the available working time before an editorial deadline and offered to draft a note to the editor.
However, in another situation, Duet failed to appreciate the nuances of an article publication timeline and repeatedly nudged about deadlines without accounting for message context or time zone differences. When drafting a Slack message, it accurately captured the author's writing tone but introduced an uncharacteristic abbreviation. From the reviewer's perspective, Duet proved valuable as a secondary safeguard against overlooked tasks, but still lacks human understanding of social dynamics and workplace relationships.
This insight applies equally to Dots. The more data an agent can see, the higher the likelihood of spotting overlooked connections across documents. Simultaneously, the risk of drawing flawed conclusions from incomplete context increases. Without an identical independent benchmark, neither platform can be deemed inherently more reliable.
Which Agent Fits Which Task
If your primary challenge is managing an overload of meetings, emails, and small commitments, Duet's focus is compelling. The product is fundamentally designed around morning briefings, situational nudges, and deciding when to alert the user. The Verge's hands-on review shows that these alerts can already be useful, even if urgency filtering still needs refinement.
If you need to sustain long-term work across multiple projects, Dots offers a more attractive framework: maintaining context across sessions, researching updates, assembling assets, and connecting specialized tooling. Scenarios involving repetitive research, engineering, and content generation are particularly promising. This remains an assessment of declared architectures rather than proof of superiority in speed or output quality.
For sensitive correspondence or payment-related actions, both platforms require careful scrutiny of connected services, granted permissions, and audit logs. Anthropic isolates credentials from the core model and enforces sensitive data leak safeguards; OpenAI relies on proprietary access rules, pre-action verification, and background restrictions. Neither system guarantees the complete absence of bugs or security risks.
Why the Industry Is Moving Toward "Always-On" AI
The arrival of Dots and Duet reflects a broader industry shift from conversational chatbots to autonomous agents equipped with memory, toolsets, and authorization to execute real work. Office productivity suites, coding assistants, and workflow automation services are all following this trajectory. The defining competitive moat is becoming the ability to retain context over days and weeks, while safely recognizing the precise moments when a human needs to intervene.
However, the market is still navigating the trust-building phase. Daily summaries lose their value if they create false alarms. Generating a report autonomously saves no time if every claim requires manual verification. For this reason, agentic systems should be evaluated on completed tasks, required corrections, and unintended side effects, rather than raw model benchmark scores.
Conclusion: Which to Choose Today
As of November 15, 2026, Claude Duet stands out as a focused assistant for managing the workday, while ChatGPT Dots provides a broader framework for sustaining multi-stage projects. In Duet's case, published hands-on testing highlights both genuine utility and frustrating edge cases. For Dots, arguments currently rest primarily on official documentation and announced capabilities. Declaring one system definitively superior is premature.
Users are best served by starting with a restricted set of connected apps and a single, easily verifiable workflow: for instance, generating daily briefings without granting email sending permissions. From there, you can assess accuracy, false alarm rates, verification time, and limit consumption. These factors will ultimately determine whether an agent becomes an asset or extra work.
Sources
- Anthropic: Introducing Claude Duet, 21.10.2026
- Anthropic: Get started with Claude Duet
- Anthropic: Claude Duet safety, privacy, and security, 23.10.2026
- OpenAI: Introducing dots, 29.09.2026; updated 03.11.2026
- OpenAI: Getting started with your dot
- OpenAI: Dots privacy, security, and safety FAQs
- The Verge: Claude Duet hands-on review, 13.11.2026
