Claude Duet and the case for choosing your agent.

What interests me is the shift in the relationship. An employee may no longer have to remember to ask for a summary of every project or check each app for the next thing that needs doing. An agent could help notice a missed customer follow-up, prepare for a meeting, or gather the pieces of a task before a person sits down to review it. Those moments are small on their own, but they are where a lot of the day goes.

I would encourage people to try Duet when they get access, with a task they already understand well. Let it prepare for a meeting or organize a follow-up, then ask what it noticed, what it missed, and whether the result actually saved anyone time. Anthropic says it may still misread a thread or get the urgency wrong. A useful trial should leave an employee close enough to catch that, especially when the agent can see email, calendars, or business systems.

Duet also arrives in a crowded group of personal agents. Our team has been trying tools such as Grok Bot, Muse, Instinct, and Dots, and Duet is another one to compare. I do not think a company should choose its entire way of working around whichever agent has the most exciting launch this week. One may be better at finding information across your existing apps. Another may be easier to talk through a problem with. A third may fit a tightly defined workflow with clearer controls. The useful choice depends on the work, the information it can reach, the quality of the result, and how comfortably employees can review it.

That is what being model and provider agnostic means to us. We want a team to understand which tools solve which problems and to be able to change its choice when the tools improve. Ruby starts by learning the process and the people who own it, then helps employees test the agents available to them against the same real task. We look at the output, the permissions needed, the human review point, and whether the team would use the new process again. The tool should earn its place in the workflow.

Spend belongs in that comparison too. Anthropic says Duet shares the usage limits on a user's Claude plan, and its background checks count toward them even when the person is not actively chatting. Larger tasks can take more model turns. Other agents have their own subscriptions, credits, or API charges, so a company should not treat every bill as if it were the same kind of token meter. Look at the total cost of getting a useful result, along with the time it gives back and the review it still requires. If an agent checks six apps all day but rarely helps someone make a decision, its activity is not the same as value.