Anthropic just quietly did something that matters a lot more than another benchmark post. In the span of a few weeks in early 2026, Claude showed up inside Chrome’s developer console, inside Excel pivot tables, inside PowerPoint decks, and as a persistent desktop agent with access to your local files. Not as a chatbot you switch to. As a thing that lives where your work actually happens. That’s a different bet — and it’s worth understanding exactly what’s been shipped, what it can do right now, and where it’s still rough around the edges.
What “Claude in Your Tools” Actually Means
There’s a pattern worth naming here. Most AI products in 2025 were tabs. You’d copy something, paste it into Claude.ai or ChatGPT, get an answer, paste it back. Friction everywhere. What’s shifted in early 2026 is that Anthropic has been systematically closing that gap — putting Claude inside the environments where work is already happening, with enough context to do something useful there.
The Claude Chrome extension reads your browser’s console errors, DOM structure, and network requests in real time. That means when a JavaScript error fires on a page you’re debugging, Claude can see what you’re seeing. It’s not you describing the problem — it’s Claude reading the actual stack trace and network tab alongside you. For a developer, that’s a different category of useful than a chat window.
Claude in Excel and Claude in PowerPoint work as add-ins. In Excel, that means Claude can help with pivot table construction and conditional formatting logic — with Opus 4.6 doing the reasoning, which, per Anthropic’s own Frontier Red Team work, is their sharpest model for structured, logic-heavy tasks. In PowerPoint, it’s available as an add-in, though the depth of slide generation and editing capabilities is still in earlier stages compared to the Excel and Chrome integrations.
None of this is magic. But the trend is clear: Claude is moving from a destination to an ingredient. And for knowledge workers, that distinction matters more than any benchmark number.
Claude Cowork: The Desktop Agent That Vibe-Works With You
The most ambitious piece of this expansion is Claude Cowork, which launched in research preview at the end of January 2026. It’s a desktop app — macOS first — that runs Claude in an isolated VM on your local machine, with full access to your local files and MCP integrations.
The framing Anthropic is using is telling. Scott White, Anthropic’s Head of Product for Enterprise, described the shift as “transitioning into vibe working” — knowledge workers directing AI rather than coding themselves. It’s a deliberate echo of Andrej Karpathy’s “vibe coding” framing, extended beyond software into any kind of desk work. Legal analysis, financial modeling, HR workflows, operations tasks — Cowork ships with domain-specific plugins for all of these.
A few things make Cowork structurally different from a browser-based chatbot:
- Persistent agent thread: For Pro and Max plan users, the agent thread persists across mobile and desktop. You can start something on your phone, continue it on your laptop, and Claude maintains context across the session.
- Local file access: Claude can read and work with files on your actual machine, not just uploads. Combined with MCP integrations, this means Claude can pull context from your local codebase, documents, or data without you having to manually feed it everything.
- Isolated VM: It runs in a sandboxed environment on your computer, which is Anthropic’s answer to the obvious concern about giving an AI agent access to your local files. Whether that’s sufficient for enterprise security requirements is a fair question — this is still a research preview.
Anthropic built Cowork itself using Claude Code in 10 days, which is either a compelling demonstration of the toolchain or a useful reminder that “built with AI in 10 days” has a wide range of quality implications depending on what you’re building. The fact that Anthropic’s own engineers now use Claude for roughly 60% of their work and are shipping 60-100 internal releases per day suggests the internal feedback loop is tight. Whether the broader product reflects that is something users are still figuring out in research preview.
Claude Code: The Quiet Workhorse Getting Serious
While Cowork gets the demo-friendly headlines, Claude Code has been getting a steady stream of substantive upgrades that deserve attention from anyone building with or on top of Claude.
The Skills API — sometimes called Agent Skills — is the most structural addition. It lets you organize capabilities into folders with SKILL.md files that tell Claude how to use them. Anthropic ships pre-built skills for PPTX, XLSX, DOCX, and PDF manipulation. In practice, this means you can point Claude Code at a folder of financial reports, give it the XLSX skill, and have it extract, restructure, and output data without writing the plumbing code yourself. It’s closer to assembling capabilities than programming them.
Other recent Claude Code additions worth knowing:
- 1M token context window for Opus 4.6 by default on Max, Team, and Enterprise plans — relevant for codebases or document sets that have historically been too large to analyze in one pass
- –bare flag for scripted automation, which makes Claude Code more useful in CI/CD pipelines and scheduled workflows
- –channels permission relay (research preview) for more granular control over what an agent can and can’t do in multi-step tasks
- Voice mode and worktrees support
- Claude Code included in every Team plan standard seat — no separate add-on pricing
The Frontier Red Team finding 500+ vulnerabilities in production open-source code using Opus 4.6 is worth pausing on. That’s not a marketing claim — it’s Anthropic’s own security team using their own model against real code and documenting what it found. It suggests that for structured, systematic analysis tasks — the kind where you want exhaustive coverage rather than creative output — Opus 4.6 is doing something genuinely useful at scale.
The Model Situation: What You’re Actually Getting
One thing that affects all of this: Anthropic has cleaned up its model lineup in a way that simplifies decisions but also removes some options.
Opus 4 and 4.1 have been removed from the model selector. Claude Sonnet 4.6 launched February 17, 2026 — same price as 4.5, meaningfully better performance, and 1M token context in beta. Opus 4.6 is the most capable model, with the full 1M context window, improved coding, and the benchmark performance on legal, financial, and coding tasks. For the office tool integrations specifically, it’s worth knowing which model you’re actually getting in each context — Anthropic uses Opus 4.6 in the Excel integration, but not every Claude touchpoint defaults to the most capable model.
| Integration | What It Does | Access | Maturity |
|---|