Claude.ai in Early 2026: What Actually Changed
In the first months of 2026, Claude.ai evolved from a chat interface into a genuine work platform. Here is what actually makes a difference in daily use, not as a feature list, but from hands-on experience.
Cowork: Claude works on your machine
In January 2026, Anthropic released Cowork . The idea: Claude gets access to your local files and works with them while you do something else. Not as a chat, but as an agent that completes tasks.
In practice, that means you tell Claude to tidy up your downloads folder, sort receipts or build a report from a CSV file. Claude reads the file names, identifies the content and categorises things on its own. An old file called IMG_4521.png? Claude recognises it as a Stripe invoice and handles it accordingly.
Since February, there is a plugin marketplace with over 38 connectors : Gmail, Google Drive, Notion, Slack, Microsoft 365. You can ask Claude to analyse data from Excel and build a PowerPoint from it, without restarting the task.
Since March, there is Dispatch , a kind of remote control. You start a task on your laptop and check the progress from your phone. It does not always work reliably yet, but the direction is right.
What stands out most: scheduled tasks. You describe once what should happen, and Claude executes it regularly. Tidy up the downloads folder every Friday. Summarise calendar entries every Monday. This is no longer a chatbot. It is a digital colleague with a calendar.
Interactive visualisations right in the chat
In mid-March, Anthropic rolled out a feature that sounds minor but noticeably changes daily work: Claude now builds diagrams, charts and interactive widgets directly in the conversation.
No side panel, no separate window. You ask for an explanation, and Claude draws a flowchart. Or you enter revenue figures, and Claude builds an interactive chart with sliders that you can adjust immediately.
Under the hood: HTML and SVG, not generated images
The visualisations load quickly, support zooming and are interactive. You can click elements, move sliders, switch views. Then ask Claude to adjust or add something.
The feature runs as a beta for all users, including the free tier. Limitation: it does not work on mobile yet, only on desktop and web. For more complex visualisations, Anthropic recommends the Opus model.
Skills: Claude learns how you work
Skills have been available since October 2025. But only now is it clear how strong the concept is. You can teach Claude repeatable workflows that persist across sessions.
For meeting notes in Confluence format, for presentations in your company CI, for text that sounds like you rather than AI. Skills are essentially instructions that Claude automatically applies to specific tasks.
Skill Creator
Built-in tool for creating your own skills directly in Claude.ai. No technical knowledge required.
Agent Skills Standard
Open standard for cross-platform skills. Create once, use everywhere.
Team Management
Since early 2026, Teams and Enterprise customers can manage skills organisation-wide and access a partner directory.
If you want to build your own skills , you can use the built-in Skill Creator or write the file yourself. The partner skills directory is growing, and the open "Agent Skills" standard ensures that skills also work outside of Claude.
Memory: Claude forgets less
Claude now has a memory. Not perfect, but useful. From past conversations, Claude derives context information: what you work on, which projects are running, what preferences you have.
This has been available to all users since early 2026, including the free tier. On top of that, there is chat search : you can search past conversations, and Claude automatically includes the context.
Not true long-term memory yet
Memory updates in the background, not in real time. If you delete conversations, the derived information disappears as well. But it is enough to avoid starting from scratch every session.
Claude in Chrome: browser agent
The Chrome extension had been in the works for a while and is now available to all paying users. Claude can read web pages, detect console errors, inspect the DOM and analyse network requests.
For developers, that means you build something with Claude Code in the terminal, and Claude tests it in the browser. Errors are detected and debugged right there.
For everyone else: you can show Claude a workflow in the browser, and Claude repeats it autonomously. This opens the path to browser automation without programming skills.
The new models: Opus 4.6 and Sonnet 4.6
Claude Opus 4.6 is currently the most capable model in the Claude family. It dominates benchmarks in software engineering and complex reasoning. In an internal test at Anthropic, Opus 4.6 outperformed every human candidate on a two-hour programming task.
Sonnet 4.6 is the faster variant with a context window of up to one million tokens in beta. Both models are available via the API and consumer apps. For daily work, that means: Opus for the hard problems, Sonnet for everything that needs to be fast.
What this means for daily work
In the past three months, Claude.ai has evolved from a good chatbot into a genuine work platform. Skills, Memory, Cowork, visualisations, the Chrome extension: each of these features is useful on its own. Together, they change how you work with AI.
You no longer just chat. You delegate. You automate. You build a system that knows you and works across sessions.
| Feature | Available since | Plan | Maturity |
|---|---|---|---|
| Cowork | January 2026 | Pro, Team, Enterprise | Stable, Dispatch still beta |
| Visualisations | March 2026 | All (including free) | Beta, desktop/web only |
| Skills | October 2025 | All | Stable, team features new |
| Memory | Early 2026 | All (including free) | Useful, not real-time |
| Chrome Extension | Early 2026 | Pro, Team, Enterprise | Stable |
| Opus 4.6 / Sonnet 4.6 | Early 2026 | All (API + Apps) | Stable |
Whether this is relevant to you depends on how you work. If you regularly create documents, analyse data, write code or process meetings, it is worth trying these new features seriously. Not as a novelty, but as a tool.
Further Reading
Frequently Asked Questions
Claude Cowork is an agent mode where Claude accesses your local files and completes tasks autonomously. You start a task, and Claude works in the background. Since February 2026, there are over 38 connectors for Gmail, Google Drive, Notion, Slack and Microsoft 365.
Skills are repeatable workflows that Claude automatically applies to specific tasks. For example, meeting notes in your company format or text that sounds like you rather than AI. Skills persist across sessions and can be shared organisation-wide since early 2026.
In early 2026, Anthropic released Opus 4.6 and Sonnet 4.6. Opus 4.6 is the most capable model and dominates benchmarks in software engineering and reasoning. Sonnet 4.6 is faster and offers a context window of up to one million tokens in beta.
The Chrome extension allows Claude to read web pages, detect console errors, inspect the DOM and analyse network requests. For developers, Claude tests code directly in the browser. For everyone else, you can show Claude a browser workflow and have it automated.
Memory is useful but not yet perfect. Claude derives context from past conversations and remembers your preferences and projects. Updates happen in the background, not in real time. If you delete conversations, the derived information disappears as well.
Yes, the visualisation feature runs as a beta for all users, including the free plan. Claude builds diagrams and interactive widgets directly in the chat using HTML and SVG. Limitation: the feature currently works only on desktop and web, not on mobile.