Before trying Claude Duet, define what deserves your attention

A useful morning brief should help you make a decision. It should tell you which client needs an answer, which deadline depends on missing information, and which issue can wait. A long summary of everything that happened can create another reading task instead.

Anthropic announced Claude Duet on October 21, 2026, as a separate proactive-agent research preview, initially waitlisted for US Pro and Max subscribers. It can prepare work across connected apps and continue while you are away. Access is not guaranteed by having a subscription. [1]

For consultants, marketers, and small teams, the useful first experiment is a tightly scoped client-attention brief. The process below is an AI Teachers proposal, not a tested Duet implementation. You can prepare the decision rules before receiving access.

1. Pick one responsibility

Choose a specific area you already own, such as keeping an approved campaign on track. Avoid a brief that covers sales, delivery, finance, and personal errands at once. You need enough familiarity with the work to spot both a misleading alert and a missing one.

Write a simple finish line: one short brief that identifies decisions you need to make today, with evidence and a proposed next step. Name the person who will review it when you are unavailable.

2. Define urgency with examples

Create three categories: decide today, review this week, and information only. Explain what qualifies for each. A question blocking an agreed launch might belong in the first category. A speculative idea without an owner probably does not.

Add a few difficult examples. A client writing “urgent” does not establish a deadline you agreed to. An old message resurfacing does not necessarily create a new problem. A missing attachment may matter more than a lengthy conversation. These examples give your review something concrete to compare against.

3. Keep the input boundary clear

List the projects, accounts, and date range the exercise should cover. Use only material your organization permits you to connect. Identify a source of truth for commitments, since a calendar event, a sales conversation, and an approved project plan can disagree.

Ask for a visible statement of missing inputs. If an assistant cannot access the current plan, it should not quietly replace that evidence with a guess. Start with one project and expand only after reviewing the results.

4. Write the brief specification

Use this as a starting instruction:

“Prepare a client-attention brief for these approved projects and this date range. Group items into decide today, review this week, and information only, using my definitions. For every item, give the source, relevant date, reason it matters, and proposed next step. Separate confirmed commitments from interpretations. Flag conflicting evidence and missing inputs. Do not invent urgency, owners, or deadlines. Draft any suggested reply separately. Do not send messages, change schedules, or update records.”

Treat that instruction as a working specification. Confirm the product’s actual controls and behavior before relying on it. A sentence in a prompt does not replace checking connected permissions and reviewing the output.

5. Run a human comparison

Prepare your own brief for the same period before reading the AI version. Compare which decisions each identified. Record missed blockers, false alarms, unsupported assumptions, and items that were correctly deferred.

Review the proposed next steps individually. A sensible summary can still recommend an inappropriate commitment. Keep the original source close to each item so checking does not become a separate research project.

6. Set realistic operating expectations

Anthropic says Duet checks connected accounts every few hours, and background work consumes the same plan allowance as other Claude use. It is not a real-time monitoring guarantee. The preview can misjudge urgency, and it cannot yet show its browser. [1]

For this exercise, keep a separate process for genuinely time-critical work. Track the effort spent preparing inputs, checking the brief, and correcting it. Compare that effort with the decisions it helped you make. Do not count a notification as a useful result by itself.

Finish with a brief you trust for a defined purpose

After several reviewed examples, decide whether to continue, narrow the scope, or stop. Keep your urgency definitions and failure examples even if you choose another tool later; they describe how your business works.

An AI Teachers readiness tutorial can help you create that specification now. Once you have access, a working session can help you evaluate a real client workflow and leave with a reviewed brief, clear boundaries, and a practical decision about whether the pilot deserves a place in your week.