Claude for Operators — Issue 005

Something unusual happened this week. Claude became the number one app in the Apple App Store, ahead of ChatGPT, ahead of Instagram, ahead of everything.

The backstory: Defence Secretary Hegseth demanded Anthropic remove AI safety guardrails from a $200 million Pentagon contract. CEO Dario Amodei refused. The administration responded by designating Anthropic a "supply chain risk to national security" and ordering federal agencies to stop using Claude. On 9 March, Anthropic sued the Trump administration in two jurisdictions.

Public reaction was immediate and one-directional. ChatGPT uninstalls spiked 295%. Claude daily sign-ups broke records every day of the week. Free users are up 60% since January. Paid subscribers have more than doubled.

We covered the Pentagon blacklisting in Quick Hits last issue. What we did not anticipate was the scale of the consumer response. This is no longer a niche story about government contracts. It is a brand event with practical implications for operators.

First, availability. Claude remains fully available through Microsoft Azure, Google Cloud, and Amazon Bedrock. Enterprise contracts are unaffected. If your company uses Claude through any of these channels, nothing changes. If you access Claude directly through claude.ai or Claude Desktop, nothing changes either. The designation applies specifically to federal agencies, not commercial customers.

Second, rate limits. More users means more pressure on infrastructure. The 2 March outage — 14 hours of global downtime — happened against this backdrop of "unprecedented demand." If your team relies on Claude for daily operations, you should have a contingency. We cover this below.

Third, opportunity. Millions of people are trying Claude for the first time this week. Many of them are knowledge workers — exactly the people who land on a tool like this and wonder "what should I actually use this for?" If you have been building Claude into your workflows for weeks or months, you are ahead. If someone on your team just downloaded it, point them to our First 30 Minutes guide.

The Pentagon story will play out in courts over months. For operators, the practical takeaway is simpler: Claude is not going anywhere, your access is not at risk, and there has never been a better time to be good at using it.

Also worth a mention for enterprise operators: Microsoft announced Copilot Cowork on 9 March — a cloud-based version of Claude's Cowork agent built into Microsoft 365. It runs in your M365 tenant, can build presentations, pull data into Excel, email colleagues, and set up meetings as background tasks. It is in limited research preview now, with broader availability later this month as part of a new M365 E7 tier ($99/user/month, launching May). If your organisation runs on Microsoft, this is worth watching closely.

On 2 March, Claude went down globally for over fourteen hours. Web, mobile, and Cowork were all affected. The API stayed largely functional, but if your workflows run through Claude Desktop or claude.ai, they stopped.

This was not the first outage. It will not be the last. If your team depends on Claude daily, you need a plan for when it is unavailable. Here is ours.

Most operators have no fallback. We have spent weeks building Claude into morning briefs, weekly reports, document drafting, data analysis. When it goes down, we go back to doing things manually — if we still remember how.

This is not a Claude-specific problem. It is the risk of building on any single AI provider.

Tier 1: Keep working (minutes)

These do not require switching tools:

Tier 2: Alternative AI (hours)

If the outage lasts more than an hour:

None of these will match Claude's output quality for most operator tasks. That is not the point. The point is keeping your day moving.

Tier 3: Process resilience (ongoing)

Build these before you need them:

We are not suggesting you stop using Claude. We are suggesting you use it like you would any critical business tool — with awareness that it can go down, and a plan for when it does.

The 2 March outage was Anthropic's longest. They attributed it to unprecedented demand. Demand is only going up. Build your fallback now, while you are not in a hurry.

Buried in this week's product updates was a significant access change: Cowork is now available to Pro plan subscribers on macOS. Previously, you needed Max ($100/month), Team, or Enterprise.

This means the $20/month Pro plan now includes Chat, Skills, Connectors, Plugins, Chrome, Claude Code, and Cowork. For solo operators who were on the fence about the upgrade from Pro to Max, the calculus just changed.

Everything. Scheduled tasks, file access, code execution, web browsing, plugin integration — it is the same Cowork as Max. The only practical difference is usage: Pro has tighter rate limits (~45 messages per 5-hour window, though Anthropic does not publish exact numbers), and Max gives you 5x or 20x that allocation.

If you are a solo operator running 2-3 Cowork tasks per day — a morning brief, some document analysis, maybe a weekly report — Pro handles it. You will hit limits if you try to run five concurrent Cowork tasks or do heavy all-day sessions.

If you have been on Pro without Cowork, here is what to do:

You will know within a day whether Pro's limits are enough for your usage pattern. If they are, you just saved $80/month.

Claude Code Review launched. On 9 March, Anthropic shipped a multi-agent code review system for Claude Code. It automatically reviews GitHub pull requests, flagging logic errors with inline comments and suggested fixes. Available for Team and Enterprise plans in research preview. Cost is $15-25 per review, scaling with PR size. If you manage an engineering team, this is worth evaluating — especially given how much AI-generated code your developers are probably already shipping.

Four Claude Code releases in one week. Versions 2.1.69 through 2.1.72 shipped between 5-10 March. Highlights: the new /loop command runs prompts on a recurring interval (e.g., /loop 5m check the deploy), cron scheduling for recurring prompts, voice mode rolling out to ~5% of users (/voice, hold spacebar), ten new speech-to-text languages (Russian, Polish, Turkish, Dutch, and more), and native MCP server management in VS Code. Full changelog on GitHub.

Haiku 3 retires on 19 April. If you or your team still use Claude Haiku 3 through the API, you have five weeks to migrate to Haiku 4.5. Sonnet 3.7 and Haiku 3.5 are already retired — API requests to those models now return errors.

Claude found 22 Firefox vulnerabilities. Anthropic partnered with Mozilla and ran Opus 4.6 against nearly 6,000 C++ files. It submitted 112 bug reports, resulting in 22 confirmed vulnerabilities — 14 classified as high-severity. Most fixes shipped in Firefox 148. The entire exercise cost $4,000 in API credits and took two weeks.

Anthropic sued the Trump administration. Following the "supply chain risk" designation on 4 March, Anthropic filed lawsuits in two jurisdictions on 9 March, arguing the designation is not legally sound. The company will continue providing models to the defence and national security community at nominal cost during transition.

HIPAA-ready Enterprise plans. Enterprise customers handling protected health information can now get a signed BAA covering chat, projects, artifacts, voice mode, connectors, enterprise search, and skills. Claude Code and Cowork are not yet covered under HIPAA.

One thing to try this week: Build your Claude failover plan.

You do not need to do anything dramatic. Take fifteen minutes. Write down your three most important Claude workflows — the ones that would hurt if they stopped working. For each one, note where the prompt lives, what data it needs, and what you would do if Claude were down for a full day.

Save it somewhere outside of Claude.

If the 2 March outage taught us anything, it is that building on AI means building with the assumption that it will occasionally not be there. The operators who handled the outage best were the ones who had already thought about it. The ones who scrambled were the ones who assumed it would always be available.

The new-user wave is real. Millions of people are discovering Claude this week. You are already ahead of them. Stay ahead by building workflows that are resilient, not just powerful.

Next week: The new connectors deep-dive continues — Google Drive, Gmail, DocuSign, and FactSet. Which ones are production-ready, and which ones need more time.

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

Just switched to Claude? Start with our Operator's Toolkit — free guides for every workflow.

Got a workflow worth sharing? Reply to this email.