Anthropic Details Evolution of Agentic Technology on Claude Platform

Anthropic previously kept memory in Claude chat separate from Claude Cowork, forcing users to repeat project details when moving from discussion to execution. A shared system is important because it turns Claude into a more continuous workplace assistant: research, planning and prior decisions made in chat can carry into Cowork tasks such as drafting management updates or conference agendas without another briefing.

Anthropic said on Aug. 25, 2026, that it was merging the two memory systems and would update saved topics while conversations are still underway. Users can inspect, edit or delete retained information. The feature is enabled by default for Free, Pro and Max subscribers across web, desktop and mobile, though iOS and Android users must install the latest app version. Sensitive details such as health, race, religion and politics are excluded by default, while government-issued identification numbers and certain other restricted data will never be stored.

Anthropic’s Claude Cowork is designed to let AI agents carry out multi-step tasks on users’ behalf, ranging from organizing information to handling enterprise workflows. Expanding beyond its original environment to the web and mobile devices could embed the AI virtual coworker more deeply in daily operations and cross-device collaboration, marking an important step in enterprise adoption of generative AI.

Anthropic has announced that Claude Cowork will expand to the web and mobile, with support for cross-platform operation and continuous cloud-based background processing. Early data show that 33.4% of users already employ the tool for business-process operations. However, available reports do not specify an exact release date, subscription price or related investment amount, and the rollout timetable and pricing plans have yet to be announced.

Anthropic positions Claude Cowork as a desktop AI workplace agent that helps companies and professionals integrate AI into their daily workflows. Expanding its usage allowance could lower barriers to experimentation and help users develop new habits. The move also reflects Anthropic’s accelerating effort to challenge Microsoft Copilot and Google Gemini.

Anthropic said it would double Claude Cowork’s usage allowance in each five-hour period for one month beginning June 5, 2026. The limited-time measure focuses on the desktop service, giving users more time and more workplace scenarios in which to test the AI agent. Anthropic hopes the initiative will further increase adoption among businesses and professionals.

Anthropic has introduced Dreams, a Claude Managed Agents feature designed to improve AI agents that operate over extended periods. As agents accumulate large volumes of task-related memories, they can encounter duplicated content, conflicting information and deteriorating decision-making performance. Dreams allows the system to proactively consolidate memories during idle periods, reducing errors and the contextual burden associated with long-term use.

Anthropic recently unveiled Dreams at a developer conference and made it available as a research preview. The company has not announced an exact date, pricing or a timeline for formal release. The feature can automatically review and reorganize an agent's memories between tasks, removing duplicate content and reconciling conflicting information. It aims to help Claude Managed Agents maintain stable performance after continuously completing multiple tasks.

Anthropic is moving Claude beyond a general-purpose chatbot and into core enterprise workflows, enabling companies to build role-based AI agents that can perform tasks. The strategy emphasizes corporate data governance, access controls and tool integration, with the goal of making Claude an intelligence layer for cross-departmental processes that works alongside existing enterprise software.

At an event on Feb. 24, Anthropic unveiled updates to Claude’s enterprise capabilities, allowing companies to create private plugin marketplaces and configure the model as an agent for specific roles. Its latest move brings Claude Managed Agents into enterprise public beta, helping organizations centrally deploy and manage AI agents. Anthropic has not disclosed pricing or contract values.

More than 500,000 lines of source code from Anthropic’s AI coding tool Claude Code could be reconstructed after a developer mistakenly included a source map in an npm package release. The leak exposed core designs for long-running AI Agent operations and context management, while also raising the risks of imitation by competitors and supply-chain attacks.

The latest disclosures show that the code includes a three-tier memory architecture for storing short-term, session and long-term information, KAIROS for autonomous operation, and a “hidden mode” for covert contribution codes. Anthropic said the incident did not affect customer data. As of April 7, security researchers had found hackers exploiting interest in the leak to distribute malware.

Anthropic launched Claude Cowork in January 2026, bringing Claude Code’s AI agent capabilities to general knowledge workers and allowing it to perform multistep tasks within designated folders. Enterprise customers account for about 80% of its business, making Cowork’s ability to integrate with existing software crucial to Anthropic’s push into the high-margin enterprise market. It also bears on investor concerns that AI could displace the software industry.

Anthropic updated Cowork on February 25, 2026, adding integrations with Google Workspace, DocuSign and FactSet and allowing enterprises to build custom plugins. On March 11, it enabled Excel and PowerPoint to share context and allowed workflows to be saved as one-click Skills. The iShares software ETF rose 9.3% over six trading sessions, while the VanEck semiconductor ETF fell 5.3% over the same period.