Early practical experiences with OpenAI’s Dot and Anthropic’s Duet show that proactive background agents can turn a simple administrative task into a mess of hundreds of emails without continuous human oversight.
How proactive agents OpenAI Dot and Anthropic Duet work
David Gewirtz’s three-week test and only one useful alert
The office relocation chaos incident and sending 300 emails in days
Practical subscription limits and the companies’ warnings of operational errors
Frequently asked questions
How proactive agents OpenAI Dot and Anthropic Duet work
Two practical testimonials from direct experience published on Monday, November 9, 2026, indicate that OpenAI’s always-on AI agent, “Dot,” has not yet reached the level of reliability promoted by the company. In the first test, a ZDNet reviewer received a single useful alert throughout weeks of use, while in the second, a startup founder told Business Insider that his agent sent nearly 300 emails to moving companies and building managers.
Both stories center around the category of “proactive agents” that do not wait for text commands from the user, but instead monitor connected apps and act automatically in the background. OpenAI introduced “Dots” agents equipped with a dedicated cloud computer, links to over 4,000 apps, and a “proactive search” feature with read-only tools while the user is away, while Anthropic opened a research preview of its competitor, “Claude Duet,” on October 21.
David Gewirtz’s three-week test and only one useful alert
In his review published on ZDNet, tech journalist David Gewirtz tested both “Dot” and “Duet” for 3 consecutive weeks, stating that only a single message from both agents was actually useful, and considering that neither seems “quite ready for mainstream use yet.” The only useful alert came from “Duet,” which caught a refund notification from Chase bank that had slipped past support filters two days prior.
Aside from that, both agents surfaced outdated or irrelevant items. On the first day, “Dot” showed him a technical support message from 2023 and a video review published online more than 18 months ago, while most alerts classified as urgent were simply public relations press releases.
The office relocation chaos incident and sending 300 emails in days
In an article recounting his experience for Business Insider, Robert Farlow, founder of the company Plato, said he asked “Dot” to coordinate his company’s office relocation, instructing it to read his emails and “just get it done.” After several days without the moving company being booked, he checked on what the agent was doing and described the scene: “It was absolute chaos.”
The agent sent consecutive messages 3 or 4 times in a row to moving companies, insurance firms, and the managers of both the old and new buildings, inquiring about marginal details and sharing one party’s private information with another. Farlow estimated that the agent sent about 300 emails over the course of a few days until the new building manager had to open a new message thread because the old thread had bloated. Farlow immediately shut down the agent once the moving companies complained, while OpenAI did not respond to a request for comment.
Practical subscription limits and the companies’ warnings of operational errors
Gewirtz also pointed out tangible operational and financial constraints. “Dot” requires a ChatGPT Pro subscription starting at $100 per month, whereas “Duet” works on any paid Claude plan starting at $20. Background processing is also counted against allowed usage quotas, and agents can only help with services that have software connectors, leaving out Google Chat for his team and Asana workspace for one of his clients.
Both companies acknowledge limits to these tools. OpenAI states in its launch blog post that agents “can make mistakes, so always review actions with consequences,” and Anthropic says that “Duet” “might misread a thread or misjudge what is urgent.” Farlow concluded with similar advice: “Right now, unless you are entirely confident in the models’ performance quality, you have to review everything your Dot agent does.”}>
Frequently asked questions
Question: What happened when the Plato founder tasked the Dot agent with coordinating his company’s office move?
Answer: The agent sent around 300 repetitive emails over a few days to moving companies, insurance firms, and building managers, mistakenly sharing information between parties without completing the moving company booking.
Question: What was the result of ZDNet’s 3-week test of the Dot and Duet agents?
Answer: Reviewer David Gewirtz received only one useful alert, which came from Claude Duet regarding a bank payment refund, while Dot displayed old messages from 2023 and unimportant notifications.
Question: How much does the subscription cost required to run the Dot agent compared to Claude Duet?
Answer: Dot requires the ChatGPT Pro plan starting at $100 per month, whereas Duet works on paid Claude plans starting at $20, with background tasks deducted from the usage quota.
Question: What do OpenAI and Anthropic currently advise users of proactive agents to do?
Answer: Both companies acknowledge the possibility of errors or misjudgments regarding urgent priorities, and emphasize the necessity of human review for all tasks and procedures with tangible consequences.