Alberta's government used Claude Code to scan 466 million lines of code in 20 hours, a task that would take 6.5 years manually. The case study signals a new category for AI in legacy system maintenance.
Alberta’s Ministry of Technology and Innovation ran 50 Claude Code agents across 27 provincial ministries in July, scanning every government-owned repository – 466 million lines of code – in about 20 hours. Anthropic published the case study on July 6, estimating a manual review would have taken 6.5 years. The pipeline used a custom rules engine to flag known patterns first, then Claude Code identified exact file paths and line numbers for each finding. Issues that traditional automated scanning tools had missed were surfaced, Anthropic said.
The same week, Anthropic launched Claude for Government Desktop in public beta, a FedRAMP High authorized environment with Claude Code and Cowork built in. U.S. federal, state, and local agencies can access it at $1 per agency per month through August 2026. The case study positions Claude Code as an autonomous security auditing platform for legacy government infrastructure, not just a coding assistant, the company said.
On July 7, Anthropic expanded Claude Cowork from desktop-only to the web and mobile (iOS and Android beta), rolling out to Max plan subscribers first. Sessions now run on Anthropic’s servers as remote agents, meaning scheduled tasks keep running even when no device is offline – a change from the previous desktop-only model where agents stopped when the app closed. Anthropic shared that over 90% of Cowork sessions are not software development tasks (dev work accounts for just 8.7% of sessions). Memo drafting, RFP review, deck creation, and case work dominate. The company doubled usage limits through August 5.
Anthropic also updated its Microsoft 365 connector on July 7 to support write operations. Claude can now draft and send emails, create and delete calendar events, and create or update files in OneDrive and SharePoint. Previously the connector was read-only. Every email sent by Claude includes an agent-attribution header. Attachments are not supported. Write tools require admin enablement via the Admin Console, and the zero-data-retention design clears most IT security gates without architectural changes, Anthropic said.
On July 9, Anthropic launched Claude Reflect in beta, a monthly recap of Claude usage at Settings > Reflect. It scores users across four dimensions: Delegation, Description, Discernment, and Diligence. The scoring model quantifies the gap Anthropic identified in its Agentic Coding Trends Report: people use AI a lot but delegate very little. Available on Free, Pro, and Max plans with Memory enabled.
OpenAI added inline visualizations in Codex tasks on July 13. Data analysis code that generates charts now renders directly in the Codex task panel. Codex Remote reached general availability, allowing users to start or continue work on a connected Mac or Windows host from an iPhone or Android, with QR code pairing. A new DigitalOcean plugin lets Codex automatically provision a Droplet, configure SSH, and register it as a remote workspace – enabling a fully cloud-native workflow without a local machine.
GitHub Copilot added Codex as an agent provider in public preview for JetBrains IDEs (IntelliJ IDEA, PyCharm, WebStorm) on July 7. Developers can run autonomous Codex agent sessions from within their JetBrains environment, including Autopilot permission mode. The expansion gives Flutter developers using Android Studio a credible agent option within their existing environment.
On June 18, Cursor acquired Continue, the open-source coding agent with 34,000 GitHub stars and 4,800 forks. Cursor gave users until July 15 to export their data before deletion. Continue shipped a final 2.0.0 release covering VS Code, CLI, and JetBrains. Cursor itself is being acquired by SpaceX for $60 billion in stock, expected to close in Q3. The acquisition absorbs the leading open-source multi-IDE architecture before SpaceX takes the helm, locking in JetBrains and CLI surface area alongside the VS Code-native Cursor product.
Flutter’s GitHub repository flutter/skills was renamed to flutter/agent-plugins on July 13. Agent Skills are markdown instruction files that give AI coding agents task-specific Flutter and Dart workflows. Place them in .agents/skills/ and the agent follows Flutter best practices autonomously. Available skills include architecting apps using the recommended layered structure, fixing layout errors, and implementing responsive layouts. Requires Dart 3.9+.
The flutter_ai_assistant package on pub.dev gives Flutter apps a drop-in AI agent that operates the UI through the Semantics tree. It plans and executes multi-step tasks without pixel coordinates. First-class support for Gemini, Claude, and OpenAI. Intended use cases include E2E test automation via natural language, AI-powered in-app support, and UI demo generation. The Semantics tree is becoming the standard interface layer between LLMs and Flutter apps, making Flutter uniquely suited for agentic UI automation compared to native iOS or Android, the package documentation states.
Drafted by a large language model from the source reporting linked above, then screened by automated publishing checks. It is not read by a journalist before publication. Some articles cite our Alpha Score. Verify prices and figures against the original source. Educational coverage, not personalized advice.