New integrations with Figma and Canva on mobile, auto mode for Claude Code and Cowork using your computer: Anthropic is pushing towards the era of truly operational AI.
Claude is going through a phase of evolution that has little to do with a simple ‘smarter chatbot’ and a lot to do with the idea of a complete operational hub for everyday work. The latest innovations announced by Anthropic all point in this direction: integrations with professional tools such as Figma, Canva and Amplitude now work directly from Claude’s mobile app, Claude Code introduces an automatic mode with an integrated security classifier, and the computer use function in Cowork allows the model to open applications, fill in spreadsheets and navigate desktops almost like a human assistant sitting at your desk.
Let’s start with mobile, the real hot front of AI use. With the latest update, users can open and navigate Figma files, create slides in Canva or consult Amplitude dashboards without ever leaving Claude’s interface on the smartphone. This is not just a ‘link’ to the tools, but a native integration: the model can read a design, comment on it, propose variants, generate copy for a presentation or interpret product and marketing charts in real time. In practice, the work pipeline goes from ‘I open Figma, then I ask the AI’ to ‘I ask the AI, which opens Figma and works on it’. For those moving between design, growth and analytics, it means drastically reducing the constant friction of jumping between apps and contexts.
The second block of novelties is more technical but equally disruptive: Claude Code introduces auto mode, an experimental mode that allows the model to decide autonomously which coding actions to perform, filtered however by a security classifier that acts as an internal guardian. Each tool call, editing a file, creating a directory, executing a script, is first assessed by a second model that estimates the risk; what appears safe is automatically executed, the rest is blocked or deferred to human judgement. The objective is clear: to push towards ‘delegated engineering‘, where the developer defines the architecture and objectives, and the AI manages the implementation semi-autonomously, reducing downtime but without opening the door to destructive commands or prompt injections hidden in the code or documentation. It is an ambitious experiment on a slippery slope: if the classifier holds up, it paves the way for increasingly automated development teams; if it fails, it will become a case study on how difficult it is to let AI really self-regulate.
The third piece completes the picture: the new computer use feature in Claude Cowork. By activating it, Claude can use your computer directly: open apps, browse the browser, populate spreadsheets, search local files, work on integrations with dozens of tools such as Notion, Gmail, Slack, Google Docs, Calendar and Git repositories. Thanks to the Dispatch mode, you can even control this ‘digital colleague’ from your smartphone while the heavy work happens on your desktop: you send a request from your phone, your computer performs the task in the background, you receive a notification when the task is completed. It is a reversal of classical logic: it is no longer you who ‘uses’ the PC with the help of the AI, it is the AI that uses the PC on your behalf, within the limits you assign to it.
In short, Claude does not want to remain a shining model confined to the chatroom, it wants to become the connective tissue that holds tools, files, flows and devices together. It’s a move that comes as the market polarises between large generalist models and small vertical agents, and Anthropic tries to play on both tables: high-level intelligence and operational agents distributed across your work stack. If it works, in a few months we won’t be talking about ‘opening Claude’ anymore, but about ‘having Claude working’, and that’s not just a linguistic nuance, it’s a redistribution of digital labour that someone, somewhere will start to feel in their paycheck too.
