Claude for Operators — Issue 010

Something shifted this week. Cowork launched as a general release on 9 April with a full enterprise feature set: role-based access controls, spend limits, usage analytics, monitoring integration (OpenTelemetry), and a Zoom connector. These are not experimental add-ons. They are the features that procurement teams, IT directors, and compliance officers ask about before they sign off on a platform. Our Admin Setup guide walks through each one. Anthropic just checked every box on the enterprise readiness list.

The investment context makes it clear this is not slowing down. The Google and Broadcom compute deal gives Anthropic access to 3.5 GW of TPU capacity, underpinning a reported $30B+ annualised revenue run rate. An IPO as early as Q4 2026 is on the table. The platform is being built to scale, and the enterprise features landing this week are proof of that trajectory.

On the same day Cowork launched, OpenAI launched a $100/month Pro plan targeting Claude Max directly. Ten times the Codex access through 31 May (dropping to five times after), clearly aimed at the power-user segment Anthropic has been winning. Competition is intensifying at the top of the market, and operators benefit from that pressure.

Here is the honest part. Four more service incidents hit between 6 and 13 April, following four in the preceding week. Elevated error rates, degraded performance, and authentication failures across Claude.ai and Claude Code. The Register ran a piece headlined "Claude is getting worse, according to Claude." That is not a great look the same week you are pitching enterprise reliability.

The tension is real. Cowork is now powerful enough to depend on. Role-based access controls mean you can roll it out to a team of 50 with granular permissions. Spend limits mean finance will not get a surprise bill. Analytics mean you can measure adoption and ROI. These are the features that turn a tool into infrastructure.

But infrastructure has to stay up. Four outages in eight days is not infrastructure-grade reliability. The API was less affected than the consumer products, which matters if you are building on the API. But most operators are not on the API. They are in Claude.ai and Cowork, where the disruptions hit hardest.

The practical takeaway: embed Claude into your operations now. The feature set is genuinely ready. But build with resilience in mind. Assume it will go down. Structure your workflows so that a two-hour outage is an inconvenience, not a crisis. The reliability will catch up. The operators who embedded early, with eyes open, will be the ones who benefit most when it does.

Claude now lives inside Microsoft Word. The add-in launched in public beta between 10 and 13 April, joining the existing Excel and PowerPoint add-ins to complete the Office suite. If your team lives in Word (and most operations teams do), this is worth paying attention to.

What it does. The add-in appears as a sidebar inside Word. You can highlight text, ask Claude to rewrite it, summarise a section, or draft new content based on the document's context. So far, that sounds like every other AI writing tool. The differentiators are in the details.

First, citations. Claude references specific sections of your document when it makes suggestions, so you can trace its reasoning back to the source material. For legal teams reviewing contracts or operations teams working through SOPs, this matters. You need to know why it suggested a change, not just what it changed.

Second, tracked changes. Claude's edits appear as tracked changes in Word's native revision system. Your existing review workflow does not need to change. Accept, reject, or modify each suggestion the same way you would handle edits from a colleague.

Third, comment threads. You can have a back-and-forth conversation about a specific section of the document, right in the sidebar. Ask Claude to explain a clause, suggest alternative language, or flag potential issues. The context stays anchored to the relevant text.

Fourth, cross-app context. This is the real unlock. The Word add-in can reference data from Excel spreadsheets and PowerPoint presentations in the same workspace. Ask Claude to update a report's financial summary based on the latest figures in your linked spreadsheet, or ensure a proposal's narrative aligns with the numbers in your deck. The three add-ins work as a suite, not three separate tools.

Who it is for. Legal teams reviewing contracts and redlines. Operations teams maintaining SOPs and process documentation. Anyone who writes proposals, reports, or briefs in Word and wants a second set of eyes that understands the full document context.

The honest take. The add-in requires a Team or Enterprise plan. Free and Pro users cannot access it. You also need a compatible version of Word: the web version, Windows M365 version 2205 or later, or Mac version 16.61 or later. Older perpetual licence versions of Word are not supported.

The timing is notable. Microsoft is restricting free Copilot access in Office apps on 15 April, pushing unlicensed users out of in-app AI assistance. Meanwhile, Copilot Cowork launches on 1 May at $99/user/month in the M365 E7 bundle. Claude's Word add-in slots into the gap: available now, at a lower price point than Copilot's enterprise tier, and backed by Opus 4.6 reasoning.

If your team already uses Claude on a Team or Enterprise plan, install the Word add-in today. It takes two minutes and immediately adds value to any document-heavy workflow. Our Document Creation guide covers the setup.

Four outages in eight days. If you depend on Claude for daily operations, you felt that. Sessions interrupted mid-task. Outputs lost. Scheduled tasks that did not run. The platform is powerful, but it is not yet dependable enough for zero-downtime workflows. Here is how to build resilience into your setup so that the next outage is an inconvenience, not a crisis.

If Claude goes down while you are mid-session, your chat history may be temporarily inaccessible. The prompts you spent twenty minutes refining are stuck behind a loading screen.

Keep a local file (a simple text document or markdown file) with your most-used prompts. Include the full prompt text, not just a description. When the service comes back, you paste and resume instead of reconstructing from memory. If you use Cowork Projects, the custom instructions persist in the project, but the session context does not survive an outage mid-conversation. Better yet, turn your best prompts into reusable Skills so they load automatically every session.

When Claude produces something you need (a report, analysis, draft, or data transformation), copy it out of the chat immediately. Do not assume the session will persist. Save it to your local filesystem, paste it into your working document, or drop it into your team's shared drive.

In Cowork, outputs saved to a linked folder are already local. But anything that exists only in the chat window is vulnerable. Make "copy the output" a habit, not an afterthought.

During the recent outages, the API was less affected than Claude.ai and Cowork. Opus and Sonnet were the first to degrade. Haiku 4.5, being lighter, often stayed available longer.

If you have API access (Team, Enterprise, or a separate API key), keep a simple script or API client configured to hit Haiku 4.5. It will not match Opus on complex reasoning, but it handles summarisation, drafting, categorisation, and formatting well enough to keep basic operations moving. The fallback does not need to be perfect. It needs to exist.

If you use Claude's scheduled tasks, design them so they can run twice without causing problems. If a scheduled task was supposed to process your inbox at 9am and the service was down, it should be safe to run it again at 11am without duplicating work or missing items.

In practice: have your tasks check what has already been processed before starting. Use date stamps in filenames. Avoid tasks that append without checking for duplicates.

Not every task needs Claude. When the service drops, some work can wait and some cannot. Decide in advance.

Write a short list: what workflows pause until Claude is back (competitive research, document drafts, data analysis), and what needs a manual fallback (client responses, time-sensitive reports, scheduled deliverables). Share it with your team so everyone knows the plan before the next incident.

Bookmark status.anthropic.com and subscribe to email or SMS notifications. Knowing the service is down before you waste ten minutes troubleshooting saves real time. The status page also shows which specific services are affected (Claude.ai, API, Claude Code), so you know whether your fallback options are available.

The platform will get more reliable. The investment in infrastructure makes that a near-certainty. But "it will get better eventually" is not an operations plan. Build the resilience now, while the incidents are frequent enough to remind you why it matters.

Full guide: Workflow Orchestration | Scheduled Tasks

Managed Agents public beta. Anthropic launched Managed Agents on 8 April, priced at $0.08 per session-hour. Launch partners include Notion, Rakuten, Asana, and Sentry. Enterprise and API only for now, but this is the clearest signal yet of where the platform is headed: persistent agents that run in the background on your behalf. Worth tracking even if you cannot access it today.

Advisor Tool API beta. Anthropic's new Advisor strategy lets a cheaper model consult Opus 4.6 only when it needs help, reducing API costs by 11.9% in benchmarks. If you run high-volume API workloads, this is a meaningful cost optimisation. The routing is automatic once enabled.

Project Glasswing. Anthropic's cybersecurity initiative uses a non-public model called Mythos Preview to find zero-day vulnerabilities in critical software. Over 40 partners, including AWS, Apple, Google, Microsoft, and CrowdStrike, are involved. Mythos Preview will not be released publicly. Worth noting for enterprise operators who need to track Anthropic's security posture, but nothing actionable for day-to-day operations today.

Haiku 3 retires 19 April. If you have any API integrations still pointing at claude-3-haiku-20240307, migrate to Haiku 4.5 now. Requests will fail after the retirement date. Our Models page has the migration details.

Sonnet 4.5 and 4.0 losing 1M context on 30 April. Still on track. After 30 April, requests exceeding the standard 200K window on these older models will return an error. Migrate to Sonnet 4.6 or Opus 4.6, which support 1M context at standard pricing. Our Models page covers the migration path.

Claude Code v2.1.93 through v2.1.101. A busy stretch of releases. Notable additions: /team-onboarding command for standardising setup across teams, OS certificate authority trust for enterprise networks, and Bedrock/Mantle support for AWS-hosted deployments. If you run Claude Code in a corporate environment with custom certificates, the CA trust update alone is worth upgrading for. Our Claude Code page has the full rundown.

One thing to do this week: Update your Cowork setup now that it is out of preview.

New features are available today: role-based permissions, Zoom connector, spend limits, and usage analytics. If you are on a Team or Enterprise plan, configure the spend limits and connector permissions before your team discovers them on their own. If you have been holding off on a broader rollout because Cowork was still in preview, that objection is gone.

While you are at it, run through the outage resilience checklist above. Save your key prompts locally. Set up status notifications. Decide which workflows can wait and which need a manual fallback. The platform is ready for serious use. Build the safety net before you need it.

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.