This week in AI: 8 things you may have missed

AI moved fast this week. Two model launches, a brand-new product from Anthropic, image generation that thinks before it draws, and a quiet shake-up to the Pro plan you might be on. Here's the 60-second catchup so you can get back to building.

THIS WEEK IN AI

8 things from Anthropic + OpenAI

ANTHROPIC

• Claude Opus 4.7 ships GA. Better long-running coding, 3× higher vision resolution, a new "xhigh" reasoning effort, and per-task budgets. Released April 16 and rolling through Pro/Max and the API.
• Claude Design launches. A new product for spinning up prototypes, slides, and one-pagers from a prompt — Anthropic's bet on "thinking visually," not just "writing about visuals." (April 17)
• Managed Agents on Claude Platform. A hosted service with stable interfaces for sessions, harnesses, and sandboxes. Built for long-horizon agent work where durable state, tool safety, and warm starts matter.
• Claude Code Pro shake-up + quality fixes. Anthropic briefly tested pulling Code from the $20 Pro plan for ~2% of new signups (existing subs unaffected), then reset usage limits and reverted three recent changes that had quietly degraded coding output.

OPENAI

• GPT-5.5 ships. OpenAI calls it the "smartest, most intuitive" model yet — better coding, computer use, and deep research. Live in Plus/Pro/Business and Codex on Thursday; explicitly framed as the foundation for OpenAI's "super app." (April 23)
• ChatGPT Images 2.0 (gpt-image-2). Native reasoning before drawing, 2K output, multi-image consistency (up to 8 coherent images from one prompt). Hit #1 on the Image Arena leaderboard within 12 hours by a +242 point margin. (April 21)
• ChatGPT Library. Every file you upload or create is auto-saved so you can find it, reuse it, and build on it later — the missing memory layer for files.
• GPT-5.4-Cyber rolls out. A defensive-cyber variant with looser permissions for vetted users, plus a new tiered-access program. Niche, but signals the start of model-by-vertical packaging.

---

FROM THE SITE

Three things to read after this recap:

• Vibe Coding 101: Build Real Apps with Claude, Cursor, and Bolt — The complete beginner's guide to the exact stack readers of this newsletter use.
• AI Meeting Notes Cleaner — Weekend build with Claude that turns rough Zoom transcripts into clean summaries with action items.
• AI Feedback Triage Widget — Embed-in-seconds widget that auto-sorts user feedback into bugs, features, and praise.

---

3 TAKEAWAYS FROM THE WEEK

• The "super app" race is real. OpenAI is openly bundling ChatGPT + Codex + browser. Anthropic is quieter, but Managed Agents + Code + Design tells the same story. Pick a stack and lean in — the toolchain is going to keep collapsing into one surface.
• Reasoning is leaking into every layer. GPT-5.5 reasons in chat. gpt-image-2 reasons before it draws. Opus 4.7 added an xhigh effort level. The differentiator is no longer "can it think?" — it's "where else can it think?"
• Plan changes are where you'll feel the squeeze first. The Pro-plan test isn't an isolated blip. Expect more re-tiering as both labs figure out who actually pays for high-reasoning compute. If you build on top of these models, price your product with that volatility in mind.

---

Sneak peek at Monday…

A weekend MVP that turns the week's AI launches into one-page "should I switch?" briefs your team can actually act on — built in 6 hours with Cursor + Claude.

---

PS — Which of these 8 was new to you? Hit reply with the number — I'm picking next week's deep dive based on what people missed.