TL;DR
The short version
Nathaniel Whittemore's first week with Claude Duet says less about features than about fit. A personal agent earns its place only if it does useful work you'd otherwise put off, without costing more setup than it saves. So far Duet does that better than his ChatGPT Dot, but it's a helper, not his main way of using AI.
His Dot is a decent place to dump context but rarely acts on its own. Duet goes looking for work, sometimes too eagerly, and its best moments were drafting replies to emails he would have starred and left for later.
Built on The AI Daily Brief episode (26 October 2026). Launch details, platforms and check cadence from Anthropic's announcement and the Claude Duet help page. His usage shares and the 90% email draft are his own estimates.
Start with how you already work
Whittemore, who hosts The AI Daily Brief, found he couldn't review Duet without first describing his whole AI setup.
Strategy is his biggest use, about 60% of his AI time by his own estimate. He talks ideas through out loud, and the model trait he cares about most is pushback that's based on something. "You don't want a yes person who's saying that every idea you have shines like gold." Claude Opus 5.5 is his current favourite for this.
For research, he has Opus and OpenAI's GPT-6 Astra each dig in, then review each other's work before anything goes on air. Astra handles copywriting and data analysis. Claude Code is where he builds. It runs his show's website pipeline, the social posts and a daily 60–90 second clip.
That setup matters for judging any new agent, because the agent has to fit into it.
Proactivity has always cost too much
He's tried proactive agents before. He ran an OpenClaw "chief of staff" agent and a standing research agent. Both got dropped. "The juice wasn't worth the squeeze."
Part of that is him. He doesn't schedule most calls and handles email by starring things and clearing his inbox every few days. Booking flights through an agent makes no sense when he always uses the same airline's site.
So his test for the new personal agents is narrow. Either they make proactive work cheap enough to be worth it, or they become one ongoing thread that holds all his context, so he stops re-explaining himself across separate chats.
Dot holds context; Duet goes looking for work
His ChatGPT Dot, OpenAI's always-on agent, is useful as a place to dump context. It isn't proactive, and he finds its personality flat. After six hours of strategy talk with Opus, he gave the whole conversation to his Dot. Its reply was essentially "This information has been saved for the future."
Claude Duet went the other way. It asked him to connect Slack, Gmail and Drive straight away, then offered a list of projects it could take on.
Duet is so biased towards proactivity that I found that it maybe overestimates how valuable that set of work is because it can be proactive about it.
Nathaniel Whittemore, The AI Daily Brief
His example was sponsor scheduling. Duet saw two sponsors offered the same date in two emails and flagged it as a conflict. It wasn't one, because he has a set of slots for each of those days. Duet reminded him about four times before he told it to drop the whole topic. Anthropic warns about this: Duet "might misread a thread or miscalibrate what's urgent."
Where it earned its keep
The win came from exactly the kind of email he'd normally star and leave for later. A few times Duet drafted a reply he could edit or approve from the chat. Some he sent as written. On the most substantive one, it got him "about 90% of the way there."
He also liked the web version more than he expected. It has separate channels on top of the main thread, interactive artifacts, and a task list. It did poorly on a brand and UI design task, which went better in Claude's own design tool.
Don't pick a side yet
Whittemore doesn't see much reason to commit to one company's agent. When something works in one, "within weeks if not days it's going to become part of all of them." If Dots feel too passive to enough people, OpenAI will change that.
For now, Duet isn't his main way of using AI. It's "an easy to set up little baby version of myself that does a bunch of stuff that I don't want to do." Whether that's a business or just a feature is still open. Claire Vo reached a similar point: she found Duet well made and then stopped opening it.
