Claude for Operators — Issue 011

One week. That is all it took for Anthropic to ship a new flagship model, launch an entirely new product category, and add cloud automations to Claude Code. If you blinked, you missed at least one of them.

Opus 4.7 landed on 16 April. The headline numbers: 87.6% on SWE-bench (a coding benchmark that matters to developers) and significantly improved vision, now handling images up to 3.75 megapixels (roughly three times the previous resolution). For operators using Claude through Chat or Cowork, the upgrade was automatic. You are already on Opus 4.7. The improvement you will notice most is sharper document reading and better handling of complex images, charts, and screenshots.

There is a catch for API users. Opus 4.7 uses a new tokeniser (the system that breaks text into chunks for processing) that produces up to 35% more tokens for the same input text. The per-token price has not changed, but effective costs rise because the same document now consumes more tokens. If you run high-volume API workflows, audit your usage after switching. For everyone on Chat and Cowork plans, this does not affect you. Your plan limits stay the same.

The day after, Claude Design launched as a research preview. Describe what you need in plain language, and Claude builds it: website prototypes, slide decks, one-pagers, design systems. Our Claude Design guide covers the full workflow. It reads your codebase and brand files to match your existing style. Export to Canva, PDF, PPTX, or HTML. Powered by Opus 4.7. Available on Pro, Max, Team, and Enterprise plans.

Figma's stock dropped 6-7% on the announcement. Mike Krieger left Figma's board days before the launch. The honest take: Claude Design is genuinely useful for operators who need quick visuals without a designer. It is not replacing Figma for design teams doing production work. But for the operator who has been putting off a client deck or a landing page because they do not have a designer on staff, this is a real unlock.

Behind all of this sits serious capital. Bloomberg reported on 14 April that Anthropic has attracted investor offers at an $800 billion valuation. The company is resisting. Revenue is north of $30B annualised. An IPO is still targeting October. The platform is being built to scale, and the feature velocity reflects that.

One other thing worth addressing directly. You may have seen the conversation about Claude "getting worse." A detailed analysis of nearly 7,000 Claude Code sessions surfaced evidence that Anthropic quietly lowered the default effort level from "high" to "medium" in early March. Fortune, VentureBeat, and Axios all covered it. The frustration is understandable, particularly the lack of transparency. If you have noticed Claude giving shorter or less thorough responses, this is likely why. The fix: set effort to "high" explicitly in your prompts or use the /effort command in Claude Code. It should not require a workaround, but it does right now.

On reliability: three more incidents hit between 14 and 16 April, continuing the rough stretch from March and early April. But things stabilised from 17 April onward. The pattern is trending in the right direction, even if it is not yet where it needs to be. If you built the resilience playbook we covered in Issue 010, you were prepared. If you did not, now is still a good time.

Claude Code Routines launched on 14 April as a research preview. This is the feature that turns Claude Code from a tool you sit in front of into one that works while you sleep.

Routines are cloud-based. They run on Anthropic's infrastructure, not your laptop. Your Mac does not need to be awake. Your VPN does not need to be connected. You define a task, set a trigger, and Claude Code executes it on schedule.

Three trigger types are available:

Usage limits scale with your plan: Pro gets 5 runs per day, Max gets 15, Team and Enterprise get 25.

The honest comparison matters here. If you already use Cowork's scheduled tasks, you know the limitation: your Mac has to be awake and Cowork has to be running. Routines remove that constraint entirely. They run in the cloud, always on, no local machine required. You set them up through the Claude Code desktop app, not the terminal. If you can describe what you want in plain language, you can build a routine.

The practical use cases for operators are compelling. A nightly routine that pulls your sales data, runs the numbers, and drops a summary into Slack by 7am. A weekly routine that compiles your team's activity across tools and generates a Monday morning brief. An API-triggered routine that generates a client report whenever your CRM marks a deal as closed. A daily routine that checks your inbox for overdue invoices and flags them in a shared spreadsheet.

The Claude Code desktop also received a complete redesign alongside Routines. Multi-session sidebar, drag-and-drop panes, integrated terminal, file editor, rebuilt diff viewer, and separate workspaces per session. If you tried Claude Code before and found the interface rough, it is worth a second look.

Routines are powerful, but they are still in research preview. Expect some friction. Start with a simple scheduled routine (a daily summary or a weekly report) before building anything mission-critical.

Full guides: Cloud Routines | Claude Code | Scheduled Tasks | Workflow Orchestration

Every operator has a version of this problem. You need a client proposal deck by Friday. Or a landing page mockup for a new service. Or a one-pager to hand the board before next week's meeting. You know what it should say. You do not have a designer, and you do not have time to learn Figma.

Claude Design is built for exactly this situation. Here is how to put it to work.

Step 1: Describe what you need in plain language.

Open Claude Design and tell it what you are building. Be specific about the purpose and audience, not the design details. "A 6-slide proposal deck for a $50K consulting engagement with a mid-market SaaS company" gives Claude more to work with than "make me a nice presentation."

Include the content you want on each slide or section. Bullet points are fine. Claude will handle layout, typography, and visual hierarchy.

Step 2: Let Claude build from your design system.

During onboarding, Claude Design reads your codebase and brand files to build a design system: your colours, fonts, logo placement, spacing conventions. Every output it creates draws from that system. If you do not have formal brand guidelines, Claude will infer a coherent style from whatever you provide. Even a logo and a colour palette is enough to get started.

Step 3: Refine through conversation.

The first version will be close but not perfect. Refine it the same way you would brief a junior designer. "Make the headline larger on slide 3." "Swap the order of sections 2 and 3." "Use a darker background for the pricing table." You can also use inline comments (click on a specific element to annotate it) or adjust custom sliders for things like visual density and whitespace.

This iterative loop is where Claude Design earns its value. You are not starting from a blank canvas. You are editing a solid first draft.

Step 4: Export and deliver.

Export options cover most operator needs:

What it does well. Internal documents, quick client materials, landing page prototypes, pitch decks, one-pagers, and anything where "good enough, fast" beats "perfect, eventually." The output quality is above what most non-designers produce in PowerPoint or Google Slides.

Where it falls short. This is a research preview. Complex multi-page layouts can be inconsistent. Photo-heavy designs need manual adjustment. And collaborative workflows (multiple designers iterating on the same file) are still Figma's territory. Claude Design is a solo tool today.

The honest comparison, for now. Canva is currently easier for template-based work (social posts, simple flyers, event invitations), and Claude Design launches with native Canva integration so you can export directly into Canva for further editing. Figma is currently better for collaborative design systems and production-quality UI work. But given how fast Anthropic is shipping, expect Claude Design to close these gaps quickly. It fills the space between "I can describe what I want" and "I need it built now, without learning a design tool," and that space is going to keep expanding.

Requires a Pro, Max, Team, or Enterprise plan.

Full guide: Claude Design | Proposals and Pitches | Document Creation | Building Tools

Opus 4.7 launched 16 April. 87.6% SWE-bench, 3.75MP vision (roughly 3x the previous resolution), new tokeniser. Same per-token API pricing, but the new tokeniser produces up to 35% more tokens for the same input. Chat and Cowork users upgraded automatically at no additional cost. Our Models page has the full specs.

Claude Design: research preview. Text-to-prototype for decks, websites, one-pagers, and design systems. Powered by Opus 4.7. Exports to Canva, PDF, PPTX, HTML. Figma stock dropped 6-7% on the news. Available on Pro, Max, Team, and Enterprise. Our Claude Design guide walks through the full setup.

Claude Code redesigned. Multi-session sidebar, drag-and-drop panes, integrated terminal, file editor, rebuilt diff viewer, separate workspaces per session. If you tried Claude Code before and bounced off the interface, revisit it now. Our Claude Code guide is updated.

Sonnet 4 and Opus 4 deprecated 14 April. Both models are now deprecated and will retire on 15 June. If you have API integrations pointing at these models, start migrating to Sonnet 4.6 or Opus 4.7. Our Models page covers the migration path.

Haiku 3 retired 19 April. API calls to Haiku 3 now fail. If anything in your stack still references claude-3-haiku-20240307, it is broken. Migrate to Haiku 4.5. Note the cost difference: Haiku 4.5 is roughly 4x the price of Haiku 3. Budget accordingly.

Sonnet 4.5 and 4.0 losing 1M context on 30 April. After 30 April, requests exceeding the standard 200K context window on these older models will return an error. If you rely on long-context workflows, migrate to Sonnet 4.6 or Opus 4.7, which support 1M context at standard pricing. Our Models page has the details.

$800B valuation offers, Anthropic resisting. Bloomberg reports multiple investor approaches at $800B. Anthropic is holding out, with revenue at $30B+ annualised and an IPO still targeting October 2026.

Outages 14-16 April, stabilised 17 April onward. Three more incidents affecting Claude.ai and Claude Code. The trend is improving after the worst stretch in March and early April, but reliability is still below where operators need it. Bookmark status.anthropic.com.

OpenAI launched ChatGPT Pro at $100/month. The new tier matches Claude Max on price and targets the same power-user segment. Includes expanded Codex access. Competition at the top of the market is heating up, and operators benefit from that pressure on both platforms. Our plan comparison and switching guide are both updated with the new pricing.

Managed Agents early results. Two weeks in, launch partners are sharing numbers. Rakuten's release cycle went from quarterly to every two weeks with 97% fewer critical errors. Sentry's agent goes from flagged bug to open pull request autonomously. Still Enterprise and API only, but the results validate the direction.

Enterprise pricing change. Anthropic removed bundled token allowances from Enterprise seats. Every token is now billed at standard API rates on top of the base seat fee. If you are on an Enterprise plan, review your billing to understand the impact.

Claude "getting worse" controversy. Fortune, VentureBeat, and Axios all covered user frustration over Claude's perceived decline. The root cause: default effort was quietly lowered from "high" to "medium" in March. No announcement, no changelog entry. If you have noticed shorter or less thorough responses, set effort to "high" explicitly. We cover the workaround in The Signal above.

One thing to do this week: Try Claude Design.

Open it up and describe a one-pager or slide deck you have been putting off. The board summary. The client proposal. The landing page for that new service. See what comes out. The output will not be portfolio quality, but it will be better than the blank page you have been staring at. Refine it through a few rounds of conversation. Export to PDF and send it.

That is the whole point. Not perfect design. Finished work, delivered on time, without waiting for a designer who is booked three weeks out.

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.