Anthropic pushed eight new ways to work with Claude into the wild between early March and mid-April 2026. Memory went free. Excel and PowerPoint started talking to each other. A phone turned into a remote control for a desktop agent. Claude Code quietly gained the ability to keep shipping work while you sleep.
The pattern is not subtle. Anthropic is wiring Claude into every surface where actual work happens: inside your spreadsheet, inside your terminal, inside your pocket, inside the folders on your Mac. Six weeks is not a roadmap. It is a land grab.
Here is what changed, who should care, and how to actually use any of it today.
On March 2, Anthropic dropped the paywall on Claude’s memory feature and rolled out an import tool that pulls your old ChatGPT conversations straight into Claude. Free users now get cross-conversation recall that sat behind a $20-a-month plan for eight months.
What it does: as you chat, Claude builds a plain-text summary of your preferences, ongoing projects, tone rules, recurring people, and files you keep referencing. That summary rides along into future conversations, so you stop repeating yourself every time you open a new chat. You can read and edit the memory file directly. It is not a black box.
Who it’s for: anyone who uses Claude more than twice a week. If you have ever typed “I am a freelance illustrator working on a children’s book called…” into three consecutive new chats, this is the upgrade you were waiting for.
Practical use cases: a product manager can let Claude remember the name of every feature in the current sprint, so status updates drop the setup paragraph. A parent using Claude for meal planning no longer re-lists allergies and dietary rules each week. A founder drafting investor updates can keep the company’s pitch framing in memory instead of pasting it every time.
How to turn it on: Settings → Personalization → Memory, then flip it on. The import tool lives in the same settings menu under “Bring your history from another chatbot”. Expect the first pass to grab about three months of ChatGPT history if you point it there.
Pro move: open the memory file once a month and delete anything outdated. A stale memory is worse than no memory. If you switched jobs in February, Claude should not still think you work at your old company in June.
Caveat: memory is per-account and does not sync across different Claude accounts you might have. For sensitive projects, start a fresh session with memory paused rather than letting it overwrite your profile.
Ten days later came custom visuals in chat. Claude now draws interactive charts and clickable diagrams directly inside a response, rendered in HTML and SVG instead of kicked out to the Artifacts panel.
What it does: when Claude decides a visual helps the explanation, it injects a working widget into the reply. Compound interest shows up as a slider-driven curve. A periodic table renders as a clickable grid where each element expands. A decision tree comes with expandable branches and dropdowns. These visuals are ephemeral. They live in the message, not in a saved file, and they can shift as the conversation evolves.
Who it’s for: everyone, because it is on by default across every plan, free accounts included. It matters most for learners, teachers, analysts, and anyone explaining a concept to a non-technical audience.
Practical use cases: a high school teacher prepping a lesson on quadratic equations can ask Claude to show how the curve moves as coefficients change, and the student can drag the slider live. A startup founder briefing investors can request a clickable flowchart of a funding scenario and adjust numbers in real time during the meeting.
How to trigger one: you do not have to. Claude produces a visual whenever it thinks one helps. If you want to force it, say something like “visualize how this changes over time” or “draw a diagram of the dependencies”.
Pro move: screenshot the visual if you want to keep it. These are inline and can disappear when the conversation scrolls or gets edited. They are not Artifacts, so asking Claude to “save this” does not behave the same way.
Caveat: the visuals are HTML/SVG, not editable files. You cannot export a compound-interest slider to a PowerPoint deck as-is. For persistent deliverables, ask Claude to create an Artifact or a .pptx instead.
Starting March 11, Claude for Excel and Claude for PowerPoint began sharing a single continuous session through a feature Anthropic calls Cross-App Context. On April 13, a Word add-in joined them, completing the classic Microsoft Office trio.
What it does: Claude can read from one Office app and write to another without losing the thread. Instructions and task history move with you between apps, so Claude is not starting fresh each time you switch tools. Anthropic also shipped prebuilt Skills, which are repeatable one-click workflows aimed first at finance teams. The starter pack includes DCF templates, LBO models, investment-banking deck reviews, and variance analyses. Teams can save their own Skills for internal reuse.
Who it’s for: office workers whose days bounce between a spreadsheet and a presentation. Finance, consulting, sales ops, investor relations, anyone who builds decks from numbers.
Practical use cases: an analyst can ask Claude to pull quarterly variance numbers from an open Excel workbook and auto-generate the executive summary slide in PowerPoint, without a single copy-paste. An IR lead preparing for an earnings call can let Claude summarize a model in Excel and push the bullet points into speaker notes in PowerPoint. A consultant can spin up a client deck from a modeling workbook in minutes instead of hours.
How to access: install the Claude for Excel and Claude for PowerPoint add-ins from Microsoft AppSource if you haven’t already. Sign in with a paid Claude account. On Mac or Windows, open both apps with the Claude sidebar visible and the context carries automatically.
Pro move: build and save Skills for the recurring work. A quarterly board deck, a monthly cohort report, a weekly pipeline review, a standing investor update: turn any of them into a one-click Skill and your whole team stops rebuilding the same thing.
Caveat: cross-app context is in beta, and the cleanest workflows today are Excel-to-Deck. Enterprise deployment is also available through Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry if your org routes through a model gateway.
Dispatch, announced as a research preview on March 17, lets you text a task to Claude from your phone and have your Mac execute it. The conversation syncs in both directions.
What it does: your phone sends text instructions. Your desktop does the actual work: opening files, running connectors, controlling your browser, processing folders. Files never leave your machine. It is one continuous conversation that follows you from couch to desk to office.
Who it’s for: people who want to kick off real work without being in front of their computer. Busy knowledge workers, parents who remember the thing they needed to do while out with the kids, consultants between meetings, anyone whose best ideas arrive on a walk.
Practical use cases: on the train home, you can tell Claude to pull the week’s Slack threads and draft a Monday briefing, so the file is waiting when you open the laptop. Waiting at school pickup, you can ask Claude to reorganize a messy Downloads folder into dated subfolders. On a Sunday night, you can queue up a weekly status report to be ready by Monday morning.
How to access: install the Claude Desktop app on your Mac and the Claude mobile app on iOS or Android. Sign in with the same Pro or Max account on both. From the mobile app, start a Cowork session and your desktop picks up.
Pro move: pair Dispatch with a clean, well-labeled folder structure on your desktop. Claude can only work with what it can find. Half the time people think Dispatch is broken, their file naming is just chaotic.
Caveat: the computer has to stay awake and the Claude app has to stay open. If your Mac sleeps, Dispatch pauses. And this is a research preview, so expect occasional hiccups on complex multi-step tasks.
Projects on Claude Cowork desktop binds a folder, its instructions, and an evolving task history to one persistent workspace. Instead of opening a new session for every assignment, you return to the same project with its context intact.
A caveat up front: coverage of Projects as of mid-April describes it as rolling out rather than fully generally available. Some users already have it, others will see it appear soon. If you don’t see it yet, check for a desktop app update.
What it does: you point a Project at a folder on your computer. You write the project-level instructions once (for example, “always cite sources, use British English, never edit files without asking”). Claude reuses that context and the running history of what you’ve done, so the tenth report on the same client is not starting from scratch.
Who it’s for: freelancers, consultants, solo operators, and small teams running repeatable work. Monthly client reports, quarterly research briefs, weekly operations summaries, ongoing writing projects.
Practical use cases: a content agency can set up one Project per client, each with that client’s style guide, brand voice, and past deliverables folder, and every new brief writes itself faster. A researcher working on a long report can keep source PDFs, drafts, and outline in one Project and stop re-uploading them weekly.
How to access: open Claude Cowork on desktop. If Projects is live for your account, you will see the option to create one from the sidebar. Point it at a folder and add instructions.
Pro move: treat the Project instructions like a standing prompt. Be explicit about format, tone, what to ignore, and when to ask for confirmation. The tighter the instructions, the less you have to restate later.
Caveat: because Projects is still being rolled out, features and UI may shift. Keep important files backed up outside the Project folder until you trust the setup.
On March 23, computer use arrived in Claude Cowork and Claude Code on Pro and Max plans, with macOS and Windows support.
What it does: when a task has no clean connector, meaning no Slack API, no Google Drive link, no CLI, Claude drives your browser, clicks through your apps, types on your keyboard, and reads your screen to complete the job. It prefers the precise tool first and falls back to GUI control only when it has to. Sensitive app categories like investment and trading platforms are blocked by default.
Who it’s for: power users who want to offload the clickwork of the modern web. Operations, admin, research, account management, anyone whose job includes “open this site, click these buttons, fill in this form, repeat”.
Practical use cases: a recruiter can let Claude log into a hiring portal and export candidate data into a local CSV. A small business owner can have Claude pull invoices from a supplier website that has no API and drop them into a folder. A real-estate agent can scrape listing sites for updates and compile a weekly comp report.
How to access: update to the latest Claude Desktop app, enable computer use in Cowork settings, and grant the OS-level permissions your machine asks for. Claude will request permission before touching any new app for the first time.
Pro move: watch the first run of any new workflow. Computer use is useful when it works and painful when it gets stuck on a modal dialog you did not know existed. Once you have seen a workflow succeed end-to-end, then let it run unattended.
Caveat: some apps are blocked by policy, and anything involving real-money transactions you should supervise. Anthropic calls this a research preview for a reason.
Claude Code Channels, shipped around March 20, connects a running local Claude Code session to Telegram or Discord. You message your coding agent from your phone and get real code back.
A caveat up front: I found solid third-party coverage of Channels (VentureBeat, community write-ups, plugin docs) but did not find a single canonical Anthropic announcement page. Everything below is grounded in that reporting, but treat specific behavior claims as things to confirm in the Claude Code release notes before you rely on them.
What it does: Channels runs as a local MCP subprocess on your machine. It listens for messages from a paired Telegram or Discord account, forwards them into your Claude Code session, and sends replies back through the chat app. Your code never leaves your machine. Pairing uses a one-time code, and only your paired account can send messages in.
Who it’s for: developers running long-lived Claude Code sessions. Long refactors, CI-triggered debug runs, test-suite builds, anything with a long tail of file edits where you do not want to be physically tethered to the terminal.
Practical use cases: you kick off a big refactor before lunch, then check progress and redirect Claude from Telegram while eating. On a weekend, you get a build-broken ping in Discord and ask Claude to investigate without reopening the laptop.
How to access: install the Channels plugin in Claude Code (research preview allowlist applies), pair your Telegram or Discord account with a one-time code, and start messaging. Your local Claude Code session has to be running.
Pro move: use Channels for long, slow tasks, the ones where you want to nudge progress without babysitting. Short, fast tasks are better done in the terminal.
Caveat: because coverage is mostly third-party, double-check current behavior against the Claude Code release notes before you rely on it for production workflows. And remember the plugin polls outward, so your local session still has to be running for anything to happen.
Auto Mode, a research preview Anthropic released in late March, is the safer alternative to --dangerously-skip-permissions. It lets Claude Code execute tasks without asking for approval on every file write and shell command, while a classifier flags anything risky and stops for your call.
What it does: instead of prompting you for permission on every action, Claude runs actions a classifier judges safe (reads, searches, scoped edits, normal build commands) and pauses only on the stuff that could damage your system. A fixed allowlist keeps non-state-modifying tools always on. You can add your own always-allow rules for specific patterns in your project.
Who it’s for: developers who trust Claude Code but got tired of the approval churn, and who do not want to run with permissions fully disabled. If you have ever hit “y” two hundred times in a single session, this is the upgrade.
Practical use cases: an overnight run of a large refactor proceeds without blocking on confirmations. A test-writing sprint where Claude iterates through dozens of files stops only when something genuinely risky shows up.
How to access: enable Auto Mode in your Claude Code settings or via the command-line flag documented in the release notes. Review the default allowlist for your project, add your own safe patterns, and start.
Pro move: commit often. Auto Mode reduces friction but does not eliminate mistakes. A clean git history is your undo button.
Caveat: Auto Mode is still a research preview, and the classifier is not perfect. For production code, repos with untrusted dependencies, and anything touching secrets or credentials, keep manual review on.
If you are on the free plan: turn on memory today. It is the single biggest quality-of-life improvement in this batch, costs nothing, and pays back after two sessions.
If your workday lives in Office: install Claude for Excel and PowerPoint, open them side by side, and rebuild one of your recurring deliverables as a Skill. The Skills payoff compounds every time you reuse one.
If you are a power user already on Pro or Max: Dispatch. Being able to start real work from your phone and finish it on your desktop changes what time of day you can be productive.
If you write code for a living: Auto Mode on your next long refactor. Pair it with commit-often discipline and it is the most tangible productivity jump of the six weeks.
The trajectory lines up. A Word add-in completed the Office trio on April 13. Projects are rolling out on desktop. Dispatch and computer use are still in research preview but tightening quickly. The next obvious moves: Android feature parity with iOS on Dispatch, a Linux build of Claude Desktop for the developer crowd, and deeper Skills marketplaces. The prebuilt finance packs landed for a reason, and verticals like healthcare, legal, and consulting are the natural next drops.
Expect Auto Mode to graduate from research preview by summer, probably with tighter defaults for enterprise environments. Expect Projects to pick up multi-user collaboration, which Cowork has been inching toward since February. And expect Anthropic to keep pushing the “Claude is everywhere work happens” thesis. This is what a company looks like when it has decided the browser chat window is not where the business ends.
Eight features in six weeks is not the work of a company looking to win a benchmark. It is the work of a company trying to own the thin layer between humans and their tools. OpenAI still has the widest mindshare. Google has the distribution advantage. Anthropic is betting that the winner of AI at work is whoever gets inside the software people already use, and nobody else is moving this fast on that bet.
If you pay for one seat, you are not using half of what shipped last month.
Sources
Anthropic — VentureBeat — TechCrunch — The New Stack — 9to5Mac
This article is AI-generated.
