Anthropic AI Updates: August 21, 2026

1. Computer Use, Skills API, and Files API Reach General Availability

Anthropic. Anthropic made computer use, the Skills API, and the Files API generally available for building production agents. Computer use now supports multi-action turns and adds a browser use tool aimed at web applications, the Skills API lets teams upload reusable expertise, and the Files API handles document management. Together the three primitives target teams moving agent prototypes into production. Source

2. Anthropic Launches Claude Academy

Anthropic. Anthropic introduced Claude Academy, an educational platform that teaches users how to work with AI effectively and safely, mirroring the company’s internal AI fluency training. The curriculum emphasizes broader mindsets over specific features, focusing on human agency, intentional decisions about when to use AI, ethical disclosure, and continuous learning. Source

3. The Claude Code Guide for Startups

Anthropic. Anthropic published a guide distilling five operating principles from interviews with more than a dozen fast-growing startups using Claude Code, including ClickHouse, Omni, Clay, and Artemis Security. The five rules are: everyone ships, automate tedium, trust but verify, build for rebuilding, and prototype-dogfood-productionize. It offers a concrete reference for teams standardizing agentic coding workflows. Source

4. monday.com Rebuilds Its Platform Around Human-Agent Collaboration

Anthropic. A customer story details how monday.com rebuilt its work management platform around human-agent collaboration, embedding agents throughout its workflows rather than adding isolated AI features. Since launching in May 2026, the platform has processed over 5 million agent interactions across tasks such as IT ticket triage, interview scheduling, and competitive intelligence. Source

5. How Slack Builds Human-Agent Teams

Anthropic. In an interview, Slack Chief Product Officer Jaime DeLanghe describes building effective human-agent teams by treating conversation history as a knowledge base and managing handoffs between humans and agents. She covers delegating clear roles to specialized agents, keeping channels public by default, and spreading adoption through demonstration. The account offers a practical model for organizations introducing agents into existing collaboration tools. Source

6. Claude Code v2.1.237 and v2.1.238 Add New Configuration Options

Anthropic. Anthropic tagged two Claude Code releases on August 20. New additions include a keybindingFlavor setting for Readline-style Ctrl+W behavior, a plugin marketplace headersHelper for minting HTTP headers, a --defer-shutdown-max-min flag for self-hosted runners, proxy authorization options via --proxy-authorization-command and --proxy-authorization-file, and a built-in Concise output style for streamlined replies. Source