Claude for Operators — Issue 008

Last issue we made the case that Claude has crossed from chatbot to operating system, drawing the comparison with OpenClaw. That argument landed. The response from operators was not "you are wrong" but "OK, so what do I actually set up first?"

Fair question. March shipped fourteen features in thirty-one days. Nobody can adopt all of them at once, and trying to is the fastest way to get nothing done. The operators getting the most from Claude right now are not the ones who tried everything. They are the ones who picked two or three features and went deep.

What is worth noting is the direction these releases point. Anthropic calls it the Enhancement Doctrine: Claude is not trying to replace your software stack. It is embedding itself inside it. Salesforce, DocuSign, Intuit, now Excel and PowerPoint. The value of learning Claude is not that it replaces your tools. It is that it makes the tools you already pay for meaningfully better.

We break down each major release in the Tool Review below: what is ready now, what is promising but early, and what needs more time. But if you want the short version: set up Cowork Projects, install the Office add-ins, and build your morning brief. That is a week's work. Everything else can wait until you are ready.

We tested every major feature Anthropic shipped in March. Here is the honest assessment, grouped by how ready each one is for your daily operations.

Ready now: These are production-ready. No research preview caveats. No 50% success rates. Set them up and they save you time immediately.

Cowork Projects. Persistent workspaces with their own folders, custom instructions, and project-specific memory. This is the feature that turns Cowork from "tool you explain things to" into "tool that already knows how you work." If you only set up one thing from March, make it this. Five minutes of configuration. Compound returns on every session after that. Full setup guide: Setting Up Persistent Workspaces.

Excel and PowerPoint shared context. Claude's Office add-ins now share conversation context across applications. Analyse a spreadsheet in Excel, switch to PowerPoint, and Claude already knows the data and the analysis you just ran. No re-uploading. No re-explaining. This is genuinely novel; no other AI tool does cross-application shared context in Office. We walk through a practical workflow in the Operator Workflow below.

1M context window (GA). Standard pricing since 13 March. No beta headers, no special access. The 600-page media limit means entire due diligence rooms, full quarter financials, and complete contract sets fit in a single conversation. The workflows that were impractical six weeks ago now just work. Details on our Models page.

Scheduled tasks plus connectors. The combination is what matters. Scheduled tasks pull live data from your connected tools (calendar, email, CRM, cloud storage) and deliver finished outputs on a cadence. If you have not set up a morning brief yet, start there. Five minutes of setup. Context waiting for you every morning. Walkthrough: Scheduled Tasks.

Promising but early: Impressive in the right context. Also early. Set expectations accordingly.

Computer Use. Claude can see your screen, click, type, and navigate applications. The concept is transformative. The execution is honest: roughly 50% success on complex tasks by Anthropic's own benchmark. Simple, well-defined tasks succeed much more often. Use it for low-stakes automation: vendor portal exports, repetitive data entry, screenshot-and-extract routines. macOS with Apple Silicon only. Research preview. Our Screen Automation guide covers when to use it versus connectors or Chrome.

Dispatch. Send tasks from your phone to your desktop Cowork session. Great concept, but it launched with a bug that caused silent failures, now fixed in v1.1.9493. If you tried it early and walked away frustrated, update and try again. Still macOS only, Pro and Max.

Interactive visuals. Inline interactive charts and diagrams in conversations. Useful for quick analysis and meeting prep. Less reliable on complex charts, and outputs are temporary (unlike Artifacts). All plans including Free. Beta. See our Visual Reporting guide for practical use cases.

Claude Code Channels. Push messages to Telegram and Discord from Claude Code sessions. Niche, but it signals where the platform is heading: Claude as a system that communicates back to you. Worth exploring if your team lives on either platform.

Context only; no action needed yet.

Marketplace. B2B marketplace launched 16 March with six enterprise partners. Zero commission. Will matter when the catalogue grows, but for most operators it is not actionable yet.

Copilot Cowork. Microsoft's product, not Anthropic's. Claude's Cowork technology inside M365 at $30/user/month. Now in the Frontier programme (broader than the initial preview, Microsoft blog). Worth watching if your organisation runs on Microsoft.

March was a platform-defining month. The features that matter most are not the flashiest. Projects and the Office add-ins will save you more time this quarter than Computer Use or Dispatch in their current state. The research previews are genuinely exciting, but they are previews. Build your workflows on what is reliable today and layer in the experimental features as they mature.

Most operators spend two to four hours assembling a board deck: pull the financials, build the charts, write the narrative, format the slides, cross-check the numbers. It is not hard work; it is assembly work. And it is exactly the kind of task where Claude's shared context shines.

Here is how to build a board-ready deck in about thirty minutes using Excel and PowerPoint together.

Open your financial spreadsheet in Excel. Open the Claude panel.

Claude walks through the data and produces a structured summary. Review it. Correct anything it got wrong. This is your chance to ensure the numbers are right before they go into slides.

Then ask for the narrative:

This forces Claude to think like a board presenter, not a data analyst. The output becomes your deck's through-line.

Open PowerPoint. If you have a previous board deck, open that file so Claude can see the format. Open the Claude panel; it already has the Excel context.

Claude generates the deck structure with the actual numbers from your Excel analysis. It will not be pixel-perfect; you will need to adjust layouts, resize charts, and apply your brand formatting. But the content, the narrative, and the data will be there.

Review each slide. The most common issues:

Then run the consistency check:

A board deck assembled in roughly thirty minutes instead of three hours. The numbers came from your actual spreadsheet. The narrative was structured around what the board needs to know. The format matches your existing template.

You still own the final review. Claude assembled it, you verify it. That is the right division of labour for anything going to a board.

The same workflow applies to:

The pattern is always the same: analyse in Excel, build in PowerPoint, let shared context carry the data between them.

Pentagon injunction granted. Judge Lin blocked the Trump administration's ban on federal agencies using Claude, ruling it was likely First Amendment retaliation for Anthropic's refusal to remove safety guardrails from the Pentagon contract (CNBC). The injunction is temporary, and full proceedings will take months, but Claude remains available to government customers for now. For operators, this de-risks the platform meaningfully. If colleagues have been asking "but what about the government thing," there is now a judicial ruling to point to. Our Security and Privacy guide covers what this means for your data.

Mythos model leaked. References to a next-generation Claude model codenamed "Mythos" surfaced through a CMS misconfiguration that exposed ~3,000 unpublished Anthropic assets (Fortune). No confirmed capabilities, no release date. Anthropic acknowledged "human error" but has not commented on the model itself. Worth watching, not worth planning around. We will cover it properly when there is an official announcement.

Anthropic eyeing IPO. Reports indicate a Q4 2026 initial public offering targeting a valuation north of $60 billion (The Information via Investing.com). For operators, this means continued aggressive investment in the platform. The shipping pace is unlikely to slow down. If Claude's pricing has been generous relative to compute costs, that may tighten post-IPO, but that is a problem for later.

Five service incidents in six days. Between 25 and 30 March, Claude experienced five separate disruptions: elevated error rates, stalled conversations, API timeouts, and intermittent authentication failures (status.anthropic.com). Most resolved within hours. The pattern is consistent with the broader trend: Anthropic is shipping faster than their infrastructure is scaling. Our failover advice from Issue 5 still stands: save key prompts externally, export critical outputs, know which tasks can tolerate downtime.

March usage promotion ended. The doubled off-peak limits that ran 13 to 28 March are over. Normal allocations have resumed across all plans. If your workflows expanded to fill the extra headroom, you will need to re-prioritise this week. Our Plan Comparison has the current limits for each tier.

Dispatch bug fixed. If you tried Dispatch early and tasks silently failed, update to v1.1.9493. The bug in v1.1.9310 has been patched. The feature itself is solid when it works, though it is still macOS only, with Pro and Max subscribers. Full setup guide on our Dispatch workflow page.

Claude Code v2.1.85 and v2.1.86 shipped. Improved MCP server reconnection logic, better handling of large file diffs, expanded /loop scheduling options, and performance improvements for voice mode. If you run Claude Code, update. Full changelogs on GitHub. Our Claude Code guide covers the highlights for operators.

MCP hits 97 million monthly downloads. The Model Context Protocol ecosystem continues to grow (The New Stack). More servers means more native integrations, which means less reliance on manual import/export workflows. Our Connector Directory tracks what is available.

One thing to try this week: Set up a Cowork Project.

Of everything that shipped in March (and there was a lot), Projects is the highest-value, lowest-risk starting point. No research preview. No 50% success rate. No beta caveats. Just a persistent workspace that remembers your context, your files, and your preferences.

Pick the workflow you run most often. The one where you spend the first few minutes re-explaining context every time. Create a project, link the folders, write the custom instructions. It takes five minutes.

Then run your next session inside it. Notice how you skip straight to the work. No preamble. No re-uploading. No "here is how we format reports." Just the task.

March was enormous. You do not need to absorb all of it this week. Start with Projects. Add the Office add-ins when you are ready. Experiment with Computer Use when you have a spare afternoon. The features are not going anywhere, but your time is.

Next issue: We are testing Computer Use workflows that actually work (dozens of experiments, narrowed down to the ones worth your time), and an updated plugin ranking for April.

Claude for Operators is an independent publication. Not affiliated with Anthropic.

Just getting started? Visit our Operator's Toolkit for free guides on every workflow.

Got a workflow worth sharing? Reply to this email.