Claude for Operators — Issue 004

In Issue 1, we covered the $285 billion stock crash triggered by Claude's plugin launch. Software companies were in freefall. The narrative was simple: AI replaces SaaS.

Three weeks later, Anthropic flipped that narrative on stage.

At The Briefing on 24 February, Anthropic CEO Dario Amodei and Salesforce CEO Marc Benioff appeared together to unveil "Claude for Agentforce 360" — a deep integration that gives Claude native access to Salesforce Data Cloud. Anthropic is the first LLM provider fully integrated within Salesforce's trust boundary, meaning Claude can work with your entire customer history without data leaving Salesforce's environment.

Similar deep-tier integrations were announced for DocuSign's Intelligent Agreement Management platform and Intuit's financial suite. In total, Anthropic announced ten strategic partnerships at The Briefing.

The message was deliberate: AI does not replace your software stack. AI enhances it. Analysts are calling it the "Enhancement Doctrine."

The market listened. Salesforce climbed 5%. Intuit rebounded 4%. DocuSign gained 5%. The SaaSpocalypse, for now, is over.

What this means for you: Anthropic is not trying to replace Salesforce, DocuSign, or Intuit. It is embedding Claude inside them. For operators, this means the tools you already use are about to get Claude-powered features — without you having to change platforms or rebuild workflows. The value of learning Claude is not that it replaces your tools. It is that it makes every tool you already use significantly better.

The Briefing also delivered a wave of product announcements — scheduled tasks, thirteen new connectors, private plugin marketplaces, department-specific plugins, and a revelation about Chinese AI labs. We cover scheduled tasks and connectors in the Tool Review and the rest in Quick Hits.

The single most practical announcement from The Briefing was not a partnership or a new model. It was this: Cowork now has scheduled tasks. You describe a task once, pick a cadence, and Claude runs it automatically — daily, weekly, hourly, whatever you need.

This is what a lot of operators were going to tools like OpenClaw for. It is now native in Cowork.

Type /schedule in a Cowork task. Describe what you want Claude to do. Pick how often: hourly, daily, weekly, weekdays only, or on demand. Claude runs it on your schedule and delivers finished outputs — reports, briefings, summaries, spreadsheets — every time.

No code. No API keys. No cron jobs. You describe the task the same way you would describe it to a colleague.

Scheduled tasks on their own are useful. Scheduled tasks combined with the thirteen new connectors announced at the same event are a different thing entirely.

Those connectors: Google Calendar, Google Drive, Gmail, DocuSign, Apollo, Clay, Outreach, SimilarWeb, MSCI, LegalZoom, FactSet, WordPress, and Harvey. The directory now has over fifty integrations — and they are all available on every plan, including Free.

When your connectors are set up, a scheduled task is not just Claude drafting something from memory. It is Claude pulling live data from your actual tools and doing something useful with it, on repeat, without you asking.

Here is one we set up in about five minutes.

Step 1: Connect your tools.

Go to Settings → Connectors. Authenticate Google Calendar, Gmail, and Slack. Each one takes about thirty seconds — OAuth login, grant permissions, done.

Step 2: Create the scheduled task.

Open a new Cowork task. Type /schedule. Then describe what you want:

Set the cadence to "Weekdays." Save.

Step 3: There is no step 3.

Tomorrow morning, when you open Claude Desktop, your brief is waiting. Every weekday after that, the same — updated with live data.

Weekly report compilation. Friday afternoons. Claude pulls data from connected spreadsheets in Drive, compiles the metrics you care about, and drafts a summary. If you have the Finance plugin installed, it knows your formatting preferences.

Competitor monitoring. Daily. Claude checks specific pages, extracts what changed, and summarises.

Meeting prep. Thirty minutes before your first meeting each day. Claude checks your calendar, finds the relevant documents in Drive, and drafts talking points.

Your computer needs to be awake. Scheduled tasks only run when your machine is on and Claude Desktop is open. If your laptop is closed at 7:30am, the morning brief runs when you open it — not at 7:30. This is the biggest limitation right now, and it matters if you need time-sensitive outputs.

Connectors need to be authenticated first. The scheduled task will fail silently if Claude cannot reach a connected tool because your OAuth session expired. Check your connector status periodically.

Start simple. Your first scheduled task should be something low-stakes — a daily brief, a weekly summary. Get a feel for how Claude interprets your instructions over multiple runs before you build anything you depend on.

Available on all paid plans. Pro, Max, Team, Enterprise. Not available on Free — you need a paid plan for Cowork.

Scheduled tasks are genuinely the most useful Cowork feature since plugins. The combination of scheduling and live connectors means Claude goes from "tool you open when you have a question" to "colleague who has your brief ready when you sit down."

The limitation around requiring an awake computer is real. Cloud-based scheduling is presumably coming, but it is not here yet. For now, it works best for operators who have a desktop machine that stays on, or who open their laptop at a consistent time each morning.

If you set up one thing from this issue, make it a morning brief. Five minutes of setup. Every workday after that, you start with context instead of scrambling for it.

We promised this one last issue. Claude in Chrome is the feature most operators have heard about but not tried — partly because the setup guidance is scattered, and partly because browser automation has real security considerations that nobody is spelling out clearly.

Here is the honest walkthrough.

A Chrome extension that runs in a side panel. You tell Claude what to do on a web page — click this, fill that, extract this data — and it does it. It reads the page structure, interacts with elements, and can work across multiple tabs.

This is not the same as Anthropic's "computer use" API, which takes screenshots and controls your mouse in a virtual environment. The Chrome extension uses DOM-based automation — it reads the page code directly. Different technology, different capabilities, different limitations.

No configuration files. No API keys. Under a minute.

Here is a workflow we have been running. Every week, we check competitor pricing pages — five competitors, three plan tiers each. It used to take twenty minutes of clicking, copying into a spreadsheet, and comparing.

First run — teach Claude what you want:

Open the first competitor's pricing page. Open the Claude side panel. Type:

Claude reads the page, extracts the data, and formats it. Move to the next tab. Repeat for each competitor.

After all five:

Record for replay. Claude in Chrome supports workflow recording. Perform the task once — navigate to each URL, tell Claude what to extract — and it saves the sequence. Next week, you replay it. Same steps, same data points, delivered comparison.

Schedule it. Set recorded workflows to run daily, weekly, or monthly. The pricing check becomes a background task that delivers results to you instead of something you remember to do.

This is the part most guides skip. Claude in Chrome can see everything visible in your browser when the side panel is open — including whatever is on the page.

Use it for:

Do not use it for:

The prompt injection risk. Malicious websites can embed hidden instructions in page content that attempt to override your commands to Claude. Anthropic has built defences, but independent testing shows roughly 1 in 9 sophisticated attacks still succeed. That number will improve. It is not zero today.

Practical safety rules:

Claude in Chrome is genuinely useful for reading-heavy, repetitive browser tasks — the kind where you are extracting data from the same sites every week. Workflow recording and scheduling make it a real time-saver once set up.

But it is a research preview. The model selection is behind the rest of the platform. The security model is good enough for trusted sites, not ready for sensitive ones. And the automation can break when websites change their layout.

Start small. Pick one repetitive browser task. Automate it on trusted, non-sensitive sites. Build confidence before expanding.

Full reference: Claude in Chrome guide in the Operator's Toolkit

Anthropic refused the Pentagon — and got blacklisted. Defense Secretary Hegseth demanded Anthropic remove AI safety guardrails from a $200 million contract. Dario Amodei refused. The Trump administration responded by designating Anthropic a supply-chain risk and ordering US agencies to stop using Claude. OpenAI picked up the contract. For operators, the practical impact on consumer and business products is zero — but it is worth understanding where Anthropic is drawing lines.

Chinese AI labs caught copying Claude. At The Briefing, Anthropic revealed that DeepSeek, Moonshot AI, and MiniMax created over 24,000 fraudulent accounts and ran 16 million interactions with Claude to train their own models. This explains the recent terms of service tightening we flagged in Issue 3.

Claude went down for three hours. On 2 March, a worldwide outage took down claude.ai, the console, and Claude Code. The API stayed up. Anthropic attributed it to "unprecedented demand" — free users up 60% since January, paid subscribers more than doubled since October. If your team relies on Claude daily, have a fallback plan.

Anthropic replaced its safety policy. The Responsible Scaling Policy was replaced with a more flexible "Frontier Safety Roadmap." Public goals rather than hard commitments. Separate from the Pentagon dispute but the timing raised eyebrows.

Private plugin marketplaces. Enterprise customers can now build and distribute custom plugins across their organisation through a centralised internal marketplace. If your company is on an Enterprise plan, this is how you scale custom plugins across teams.

Department-specific plugins. New plugins from The Briefing: private equity modelling, HR (job descriptions, offer letters), design (creative briefs), and operations (vendor proposals). These join the eleven knowledge-work plugins from Issue 1.

One thing to try this week: Set up a morning brief in Cowork.

Connect Gmail and Google Calendar (Settings → Connectors, thirty seconds each). Open a Cowork task. Type /schedule. Describe the brief you want — today's meetings, unread emails that need attention, anything urgent. Set it to weekdays.

Tomorrow morning, your brief is waiting.

Five minutes of setup. Every workday after that, you start with context instead of scrambling for it. That is what the platform shift from The Briefing actually looks like in practice — not a partnership announcement on a stage, but your morning starting fifteen minutes faster because Claude already did the prep.

Next week: A deeper look at the thirteen new connectors — which ones actually work well, which ones are half-baked, and how to connect Claude to Google Workspace without giving it the keys to your entire organisation.

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

Got a workflow worth sharing? Reply to this email.