Anthropic’s Claude Duet Enters a Crowded Agent Market

By Amara Osei

Anthropic says Claude Duet, an agent app for professionals and small businesses, offers access to more than 6,000 Claude Marketplace connectors and plugins, but remains limited to Pro and Max subscribers on a waitlist. TechCrunch characterized the launch as Anthropic’s “bet on agents” and reported that the software runs in the background, notifying a person only when an action requires human input.

What can Claude Duet actually do?

According to Anthropic, Duet can prepare a daily briefing and create and suggest dashboards from user data. It can monitor ongoing tasks and suggest actions without an open chat, while a virtual machine runs work in the background. Anthropic says Duet seeks permission for specific operations, a pattern ALM Corp likened to managing a junior employee rather than chatting.

Anthropic says Duet cannot handle errands, book tickets or show the progress of its incognito browsing. TechCrunch used Instinct to illustrate why that matters: visible browsing can reassure users that they retain control. ALM Corp said a morning-briefing customer may not care how news arrives, while a flight-booking customer would care more about the outcome than the route.

TechCrunch did not state how often briefings would arrive, what dashboard customization would be available or how detailed suggested actions might be. ALM Corp said those omissions will shape whether users consider Duet useful and efficient. TechCrunch’s report supplied only basic product information, leaving the preview’s practical limits unclear.

ALM Corp said proactive recommendations can reduce the work users perform, but poor suggestions can bury the interface in another tab to monitor. Anthropic did not say whether users could customize those recommendations. The review treated the quality and relevance of suggestions as central to an agent’s value.

How wide is its connector reach, and what are the risks?

Duet runs on Claude Opus 5.5 and draws its tools from the Claude Marketplace, according to Anthropic. ALM Corp said the more than 6,000 connectors and plugins represent a maximum, not an average or minimum, and quantity does not guarantee quality. TechCrunch named no popular or relevant tools, and the review said effective agents need users’ personal data.

TechCrunch did not address the permission risk created by connectors. ALM Corp said every added tool gives the agent more access to data and features, enlarging the potential attack surface and the chance of errors. The review called access levels and ease of revocation critical security questions that the report leaves unanswered.

How does Duet handle money?

Anthropic says Duet can make online purchases through Stripe’s Link digital wallet or Shopify’s Shop Pay, and transactions involving money require user approval. The agent cannot buy or sell securities, transfer funds between accounts or complete those actions without human authorization.

ALM Corp said that boundary permits office-supply purchases but blocks investments, a limit that may reassure owners reluctant to delegate investment decisions. Anthropic also introduced a classifier that analyzes every conversation message to prevent sensitive-information leakage. The review said default settings for Claude 3 Opus make Duet refuse requests for credit card numbers and government identification.

ALM Corp noted that the classifier applies to all messages, not only payment requests. The review said approval prompts slow transactions but keep a person involved in every purchase, creating a meaningful barrier to autonomous financial activity.

How does Duet compare with rival agents?

Meta said on its third-quarter call on Wednesday that more than one-quarter of Muse’s active users who engage in agent-assisted shopping spend more than $200 per transaction on average. ALM Corp said consumer agents already handle financial transactions. Duet takes a narrower path, requiring approval at every step and restricting acceptable actions, according to the review.

ALM Corp said new assistants were arriving almost weekly when Anthropic entered the agent market with Duet. The review listed Meta’s Muse, OpenAI’s Dots, Hark, Instinct, Grok Bot and Fo as competitors.

Meta has also announced a business version of Muse integrated with other applications, while OpenAI’s Dots offers a business feature through partnerships with professional-software providers. ALM Corp described an industry shift from assistants to agents and cited the crowded field as evidence of demand. The review said model performance, integration depth and guardrail effectiveness will separate the products.

Meta said on Wednesday that Muse already had more than 10 million users. Duet remains in a research preview behind a waitlist, and Anthropic disclosed no user count for it. ALM Corp said the two products have little scale comparability at this stage.

Who can use the research preview?

TechCrunch reported that users can find Claude Duet on iOS, Android and at Claude.ai. Anthropic limits access to Pro and Max subscribers, who must request waitlist placement; the report gave no date for broader availability or separate pricing beyond the existing subscription plans.

ALM Corp said research previews are generally beta releases whose features and limitations can change before a formal launch. The review advised users considering Duet for a regular workflow to treat the announced capabilities as a foundation rather than a fixed specification.

What should small teams test next?

For marketing and SEO teams, ALM Corp said reports and dashboards are already routine parts of the job. The review said Duet could overlap with weekly business reports if it scrapes connected tools, but its effectiveness remains unproven. ALM Corp warned that a misleading dashboard could prove far more harmful than no dashboard at all.

ALM Corp said small-business owners may value Duet because they often divide their work among sales, customer service and accounting, making a daily reminder of pending tasks an easy sell. Prospective users must assess which data sources they will connect, how often approval prompts interrupt work and which near-deadline tasks to keep out of briefings.

ALM Corp based its assessment on Anthropic’s description as reported by TechCrunch and did not test Duet itself. For Duet, ALM Corp’s decisive tests are implementation quality, user adoption and permission controls; TechCrunch gave no date for wider availability.