Why I'm Team Anthropic Right Now

It's Sunday, and I'm in your inbox. Anthropic shipped three things in four days and I think it's worth the interruption.

---

Why this is a bonus edition

I don't normally do midweek sends. But between last Tuesday and Friday, Anthropic quietly pushed out what I'd call the most coherent week any AI company has had this year. A new flagship model, a brand-new product that takes a real swing at Canva and Figma, and a full rebuild of how developers and builders actually use Claude day to day.

Here's what shipped, what it costs, and what I think you should actually do about it.

---

1. Claude Opus 4.7 is Anthropic's new flagship

Launched: Thursday, April 16 (Anthropic announcement)

What's actually new: Opus 4.7 is the version of Claude that quietly retook the "most capable model" crown from GPT-5.4. Anthropic reports a 13% lift on the previous Opus 4.6 across hard coding tasks, 87.6% on SWE-bench Verified (up from 80.8%), 94.2% on GPQA Diamond, and 64.4% on the Finance Agent benchmark, which is the current state of the art.

The upgrade most office professionals will feel isn't the benchmark numbers. It's vision. Claude now reads images up to 3.75 megapixels, more than 3x the resolution of older versions. That means you can drop a dense slide, a chemistry diagram, a contract scan, or a screenshot of a messy spreadsheet into Claude and it will actually read every cell, every label, every footnote. I tested it against a 6-page scanned PDF of a commercial lease over the weekend. Opus 4.6 missed a rent-escalation clause in a footnote. Opus 4.7 caught it and flagged it before I asked.

Pricing: $5 per million input tokens, $25 per million output tokens, unchanged from 4.6. One gotcha: Anthropic swapped in a new tokenizer, so the same text can count as up to 1.35x more tokens than before (Finout breakdown). For most chat users on Pro ($20/month) or Max, this doesn't change anything. For API users, budget a ~20% cost bump even at "identical" pricing.

Who should care: Anyone using Claude for research, document review, or slide analysis. If you're still on Opus 4.6, switch today. It's free to change in the model picker.

---

2. Claude Design: their answer to "Claude can't do visuals"

Launched: Friday, April 17 (Anthropic, TechCrunch coverage)

What it is: A whole new product from Anthropic Labs that turns a conversation into a finished deck, prototype, one-pager, or mockup. You describe what you want, Claude builds it, and you refine through inline comments and adjustment sliders instead of dragging shapes around. Think "Canva, but the canvas is a chat."

Why this is a real shot at Figma and Canva: Two things make Claude Design interesting. First, it reads your existing design system. Point it at your codebase, your brand files, or even a public site, and it will match the typography, colors, and spacing without you teaching it anything. Second, it exports to where you already work: PDF, PPTX, standalone HTML, Canva, or a shareable URL. It's not trying to be your design tool forever. It's trying to be the first draft.

The market noticed. Figma's stock dropped about 7% the afternoon Claude Design was announced (The Register).

My honest take: The common complaint about Claude for the last two years has been "great at writing, bad at anything visual." Claude Design doesn't fix that completely, but it's the first Anthropic product I've seen that takes design seriously as a first-class workflow, not a side feature. Early launch partners include Datadog and Brilliant, which is a pointed reminder that this is aimed at product and marketing teams, not hobbyists.

Pricing: Included for Claude Pro ($20/month), Max ($100 to $200/month), Team, and Enterprise subscribers. Separate usage bucket, so it doesn't eat into your chat limits. Rolling out gradually, so if you don't see it yet, it's coming.

Link: claude.ai/design

---

3. A complete rebuild of the Claude Code desktop app

Launched: Tuesday, April 14 (VentureBeat hands-on, MacRumors)

Fair warning: this one is for the builder slice of the audience. If you write code, manage developers, or run a product team that ships software, read on. If not, skip to the "why I'm team Anthropic" section.

What changed: The Claude Code desktop app got rebuilt from scratch around a different idea: you're not waiting for one task to finish, you're running several in parallel and checking in as each one lands. The new sidebar shows every active session grouped by project. Panes are drag-and-drop, so you can arrange the terminal, diff viewer, file editor, and HTML preview however you work. A new side chat shortcut (Cmd+;) lets you ask Claude a side question without polluting the main task's context.

The bigger news: Routines. Anthropic also launched Routines in research preview. A routine is a prompt plus a repo plus connectors, saved as one configuration. You can trigger it on a schedule, from an API call, or in response to a GitHub event like a new pull request. It runs on Anthropic's cloud, not your laptop. In plain English: you can now set Claude to review every PR at 2am, summarize your repo's issues every Monday, or run a nightly refactor sweep, all without opening your laptop.

Why leaders should care even if you don't code: Routines is the first serious "Claude as background worker" feature. If you manage a team, the right conversation with your lead engineer this week is: "what is the most tedious 30-minute thing we do every day that a scheduled Claude routine could just handle?" The answer is usually there.

Pricing: Included with existing Claude Code plans on Pro ($20/month), Max, Team, and Enterprise. Routines runs on daily caps that scale with your plan.

Link: claude.com/product/claude-code

---

Why I keep coming back to Claude

Here's the part I wanted to write most.

Every major AI company is now pitching a full "ecosystem." OpenAI has ChatGPT plus Atlas plus Sora plus Codex. Google has Gemini plus Workspace plus Chrome plus Stitch. Microsoft has Copilot in every surface they ship. They're all racing for the same thing: keep you inside their product all day.

Anthropic is running the same race, but the way they're running it feels different. Opus 4.7 is the brain. Claude Design is where knowledge workers sketch and ship visuals. Claude Code is where builders ship code. Skills, connectors, and the redesigned desktop are the rails connecting them. If you're a solopreneur, an office professional, or a leader with a team, this is now a legitimate end-to-end stack: think, draft, visualize, build, automate, all inside one account, with a model that genuinely belongs at the top of every benchmark it enters.

I'm not saying quit your ChatGPT or Gemini subscription. I'm saying that if you've tried Claude once in the last six months and bounced off, this is the week to give it another hour. Open claude.ai, switch the model to Opus 4.7, and try one real thing, a contract review, a slide deck draft, a weekly status email written in your voice. You'll feel the difference.

This is one of my daily-driver AI tools in 2026. It's worth the time to learn.

---

Pro Tip

---