Opus 4.7, Claude Design, Mythos, and the race with OpenAI

The newsletter has slipped a few weeks because we have been running around talking about Claude, Cowork, Claude Code and Skills. Next Monday we run our next Claude Code Bootcamp at Mesh Oslo, followed by a Claude Code meetup the same evening. This one is big.

Anthropic has shipped a whole lot since our last newsletter.

• A new model, Opus 4.7, with better instruction following and more.
• Claude Design turns a prompt into high-quality slides and prototypes.
• Managed Agents lets companies build always-on agents.
• Claude Cowork is now generally available, no longer a research preview.

Anthropic also published a 244-page model description for a model it will not sell yet. Claude Mythos Preview found a 27-year-old vulnerability in OpenBSD, the codebase famous for being the most audited open source operating system on the planet. Thousands of human security researchers missed it. An AI did not. Fascinating and scary at the same time.

OpenAI has not slowed down either. Workspace Agents launched today. Teams can now build and share custom agents inside ChatGPT, powered by Codex and running in the cloud. This is OpenAI's direct answer to Anthropic's Cowork. Chronicle watches your screen and builds a searchable memory of everything you do. And ChatGPT Images 2.0 took over every feed this week. In short, the best image model in the world, now well ahead of Nano Banana on benchmarks.

Claude Superpowers

The Superpowers plugin gives Claude Code a structured workflow from brainstorming to implementation and testing. Our tutorial covers installation, the seven-phase workflow, all 14 skills, the Visual Companion for browser-based mockups, the spec review loop, TDD with the Iron Law, and tips for getting the most out of it.

New to Claude Code? Start here. Then watch the full Superpowers tutorial.

Claude Code Bootcamp, 27 April

Our second Claude Code Bootcamp is happening next week. We are hosting it at Mesh Oslo on 27 April. A full day, hands-on programme for anyone who wants to learn Claude Code in a single day.

This is for project managers, consultants, analysts, lawyers, researchers, marketers, founders, and anyone who wants to multiply their output. All you need is a laptop and the willingness to try something new. Newsletter subscribers get 20 percent off with the code CCGRITAI03.

More than 600 people have signed up for the Claude Code meetup in the evening after the bootcamp. It is shaping up to be a big day.

AI Pulse

A bite-sized curation of this week's most important AI news.

• OpenAI launches Workspace Agents in ChatGPT. Teams build shared agents that handle complex tasks and long-running workflows. Powered by Codex, runs in the cloud. Available in Business, Enterprise, Edu and Teachers plans.
• OpenAI ships ChatGPT Images 2.0. Text that renders readably inside posters and UI, a thinking mode before drawing, up to 2K resolution. The internet cannot stop sharing outputs. Available via the gpt-image-2 API.
• OpenAI announces Chronicle. Background agents that watch your screen and build a searchable memory of everything you do. Captures are stored on-device, but other apps can access the files.
• Anthropic releases Claude Opus 4.7. Smarter at following instructions, checks its own work before responding, and can now read dense screenshots and diagrams in high resolution.
• Anthropic launches Claude Design in research preview. Describe what you want and it builds slides, wireframes and one-pagers. It reads your brand from existing files and applies it automatically.
• Anthropic ships Managed Agents in public beta. Build a custom AI agent, give it tools and instructions, and let it run in the cloud without your laptop open.
• Anthropic launches Mythos Preview, codenamed Project Glasswing. A security-focused model that finds software vulnerabilities. It found a 27-year-old bug in OpenBSD. Restricted access only.
• Claude Cowork is now generally available for all paid plans. A desktop tool that lets non-developers automate file and task management using AI. Previously in limited preview.
• Live Artifacts ship inside Claude Cowork. Build a dashboard once, and it auto-refreshes from your real files and apps with full version history.
• Meta unveils Muse Spark. It scores 58 percent on Humanity's Last Exam and is ten times more efficient than Llama 4 Maverick. The first Meta model to compete at the frontier since Llama 4.
• Cursor 3 ships with an agent-first interface. Manage multiple AI agents across your local machine, remote servers and cloud. No longer a code editor with AI on the side.
• OpenAI expands Codex to a desktop agent on macOS. It reads your screen, controls your cursor, runs background tasks and supports more than 90 plugins. A direct competitor to Claude computer use.
• Claude for Word launches on Pro and Max plans. Claude lives inside your document with clickable citations and tracked changes.
• Microsoft 365 connectors are now available on all Claude plans. Connect Claude to your Outlook, Teams, SharePoint and OneDrive.
• Google releases Gemma 4 as open source. 31 billion parameters, ranks third on Arena AI, runs locally.

Claude Code updates

Here are the highlights from the changelog since our last issue.

• Routines, always-on automation. Bundle a prompt, repos and tools into one config. Trigger on a schedule, via API or from GitHub events. Runs in the cloud without your machine.
• Monitor tool, streaming background events into your session. Tail a log, watch CI on a PR or babysit a training run. Each event lands as a new message Claude reacts to. No more bash sleep loops.
• PR autofix from your terminal. Run /autofix-pr in the CLI to turn on auto-fix for the current branch's PR. Claude watches CI and pushes fixes in the cloud until it is green.
• Desktop rebuild for multi-session workflows. Claude Code desktop now supports multiple parallel sessions and in-app development tools.
• /ultrareview, deep code review in the cloud. Runs multiple agents in parallel to analyse your branch or a specific pull request. Code review on autopilot.
• /ultraplan, cloud-based planning. Kick off plan mode in the cloud from your terminal. Claude drafts the plan while your CLI stays free. Review and comment on individual sections in the browser before execution.
• Push notifications via Remote Control. Claude can now send a notification to your phone when it needs your attention or finishes a task. Enable it in settings under Remote Control.
• Session recap when you come back. Leave a session, come back later, and Claude gives you a summary of where you left off.
• Native binary for faster startup. Claude Code now runs as a compiled binary instead of JavaScript. Noticeably faster to launch, especially on large projects.

Join the GritAI community

2025 was about learning to work with ChatGPT, Claude and Gemini. 2026 is about something else entirely. Agents. Practical AI skills and workflows, new content every week, and a live hangout with the community to learn and share what you are learning and building. Join the GritAI community.

Best, Aleksander and Simon