ChatGPT Dots Tested: Research, Sponsorship Audits, and the Limits of an Always-On Agent

Peter Yang tests dots, an always-on ChatGPT agent, on tasks in his one-person business. He reports useful research and a sponsorship issue he had missed, but also finds failures and a confusing interface. His comparison reflects a few days of personal testing, not a controlled benchmark.

In the version he shows, a pinned thread hosts his dot, named Zoe. He can customize its appearance, use voice and connect Slack; email is unavailable in his demonstration. He shows a cloud computer and a connection to his laptop, plus an activity list for tasks.

Yang sees Muse as more consumer-focused and Grok Bot as a more direct comparison for work. He values Grok Bot’s multiple collaborating bots; he says dots initially gives him one agent, with ChatGPT integration as its main advantage. Existing memory, conversations and connected apps make setup easier in his account.

For an X research task, Zoe and a Grok Bot scout return different posts. Yang cannot say which set is better. In Slack, Zoe researches comparable Maven courses and suggests changes to a friend’s course landing page.

Yang asks Zoe to assess his community and suggest improvements: organize member introductions, preserve useful knowledge and give help requests a next step. The displayed response says it sampled public channels and followed dozens of threads, rather than reading every thread. He also has it make memes, finding the personalized ones funnier than generic AI-news jokes.

For channel research, Yang asks for themes from six months of YouTube comments. He says the process took about half an hour and recommends showing workflow costs, starting tutorials from a blank setup and explaining what happens after a demo. The on-screen report describes a partial audit of 267 of 301 videos and Shorts and 1,862 comments or replies within its time window, not complete coverage.

To prepare a fitness app for launch, Yang asks Zoe to use an existing build thread, find issues and create six App Store screenshots. He describes the agent delegating work to other ChatGPT threads and reporting back, a chief-of-staff pattern rather than doing everything in the main conversation.

He asks the agent to suggest ambitious projects. Its list includes a paid product, a video workflow, a sponsorship audit, an app launch rehearsal and a community experience. Yang chooses to explore them in parallel.

In the sponsorship audit, Yang says the agent flags an apparent unbilled Riverside sponsorship and a possible $10,000 discrepancy involving Google. He presents these as issues to investigate, not verified losses. For his one-person business, surfacing possible omissions is useful because sponsorships are a major source of revenue.

The agent proposes a $99 product concept for product managers and builds a site with four interactive examples. Yang likes the improved visuals after giving feedback, but finds the copy weak and does not consider it a finished revenue-generating product. The idea and its value still need his judgment.

Zoe also reviews newsletter statistics through a browser and proposes ways to improve growth. Yang likes its ability to gather information across pages and synthesize recommendations, though the recommendations still need evaluation in context.

A newsletter drafting task based on a Riverside interview transcript encounters a tool-availability problem and fails. Yang says the failure was hard to find in the long conversation. When several workstreams run at once, the main thread can become difficult to monitor.

Yang appreciates ChatGPT’s voice and computer-use capabilities but worries that threads, Codex, Spaces, pages, sites, images and scheduled tasks make the interface increasingly complicated. An agent may help orchestrate that complexity, yet he argues the product itself should remain understandable.

His conclusion is that dots shows promise as a work assistant integrated with ChatGPT, while Grok Bot has features such as team bots that he considers ahead in this comparison. He sees an opportunity for people and agents to collaborate without overwhelming users. He says dots was available to ChatGPT Pro subscribers at the time of the video.