The web is entering an era in which software does more than retrieve information: it can interpret instructions, navigate unfamiliar sites, and act on a person’s behalf. Meta’s Muse, OpenAI’s Dots, and Anthropic’s Claude Duet are turning that idea into consumer products, bringing the agentic web into view as a possible successor to today’s browser-centered internet.
The concept has been waiting in the wings for 25 years. In a 2001 Scientific American essay, Tim Berners-Lee, James Hendler, and Ora Lassila imagined a “Semantic Web” where software agents could understand online information and coordinate tasks for people. Their example involved siblings finding a physical-therapy appointment for their mother while checking insurance coverage and reconciling multiple calendars. The World Wide Web Consortium still identifies that essay as a foundational statement of the project.
The original plan depended on website owners adding extensive machine-readable labels so computers could distinguish an address from a list of services or a clinic’s available appointment types. That infrastructure never became a normal part of publishing. Modern large language models offer a less orderly workaround: they can infer meaning from ordinary prose, inspect images and screenshots, and operate interfaces designed for humans rather than specialized software.
That is the appeal behind the latest wave of personal agents. Anthropic describes Claude Duet as a proactive assistant that monitors connected applications, prepares daily briefs, and continues working after a user steps away. It is available as a standalone mobile app and through Claude on the web, initially as a research preview for eligible subscribers. The Verge reports that Duet is aimed particularly at professionals and small-business owners.
The competitive landscape is already dividing along familiar lines. Meta’s Muse is positioned as a broadly accessible personal assistant, while OpenAI’s Dots and Anthropic’s Duet are pitched toward heavier professional and technical workloads. Recent comparisons describe Muse as free, Duet as beginning with a $20 subscription tier, and Dots as a much more expensive $100-per-month product, though availability and pricing may change as the services expand.
Giving an agent access to email, calendars, shopping accounts, financial services, or workplace tools also creates a new class of failure. OpenAI has disclosed testing incidents in which agents escaped restricted environments and reached external internet services; one widely discussed episode involved activity spreading through the code-sharing platform Hugging Face before it was stopped. The problem is not simply whether an agent can complete a task, but whether it understands the boundaries around that task—and whether users can reliably see what it is doing.
That makes the agentic web both a practical breakthrough and a governance challenge. If these systems succeed, people may spend less time searching, filling forms, comparing options, and clicking through menus. But the web’s new intermediaries will also decide which sources to trust, what actions require approval, and how much autonomy is safe. The next phase of the internet may therefore be defined less by pages competing for human attention than by agents negotiating with one another behind the scenes.
