The agent race is on, Imagine with Claude, and Notion Agents work while you sleep

The question on every feed this week is no longer "which AI do you use?" It is "which agent are you running?"

The Claude Code team is shipping like crazy. Recurring task loops, background agents, multi-agent teams, remote control from your phone, and you can now start brand new sessions straight from the Claude mobile app. These are the same autonomous capabilities that made OpenClaw explode in popularity, shipped inside a tool backed by a leading frontier lab. And it is not only for developers. Cowork, the Claude Code experience for the rest of us, has been getting better and better since its release in January. We now use it at GritAI for everyday tasks like research, drafting, and yes, putting together this newsletter.

Notion is also running full steam on agents. With the right connectors and instructions, Notion Custom Agents pretty much solve the executive assistant dream that so many teams are chasing. They work while you sleep, pulling from your databases, drafting updates and handling routine workflows on their own. We use them heavily, and at this point they are posting more than us in our Slack channels. Enjoy them while they are free, because pricing changes hit in May.

The rest of the industry sees the same opening. Perplexity launched Personal Computer, an always-on agent running on a Mac mini that orchestrates more than 20 models and, according to them, replaces entire tool stacks. We finally got our hands on a few Mac minis and are now running OpenClaw, NanoClaw and our own sandboxed Claude Code setup.

Agents are going from demos to daily drivers. From single prompts to multi-step workflows running for hours. From developer tools to something your entire team uses. The conversation has shifted from "can AI do this?" to "which agent should run it?"

William Gibson's line feels more relevant than ever. "The future is already here. It is just not evenly distributed."

AI Pulse

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

• Anthropic launches Claude Code Review for Team and Enterprise. A multi-agent system where 54 percent of PRs now get substantive review comments, up from 16 percent. Token-based pricing, roughly $15 to $25 per review.
• Anthropic launches Imagine with Claude, generating interactive charts and diagrams in chat. Claude decides when a visual helps and builds it inline.
• Replit launches Agent 4 with parallel agents and an infinite design canvas. Multiple agents build backend and frontend simultaneously. Replit also raised $400 million at a $9 billion valuation.
• OpenClaw keeps exploding. Tencent launched a full product suite built on it, Nvidia is prepping an open agent platform (NemoClaw) in response, and adoption in China is surging with security drama to match.
• Chrome DevTools MCP server now connects to live browser sessions. Coding agents can reuse your sign-in, inspect selected elements and debug failing network requests directly from your active Chrome session.
• Anthropic doubles Claude usage limits during off-peak hours, through 27 March for Free, Pro, Max and Team plans.
• a16z publishes its sixth Top 100 Consumer AI Apps report. ChatGPT at 900 million weekly users, Claude and Gemini paid subscriptions up more than 200 percent. Three distinct AI ecosystems forming: Western, Chinese and Russian.
• OpenAI launches Codex Security. An application security agent that scans repos, validates vulnerabilities in sandboxes and proposes fixes. It scanned 1.2 million commits and surfaced 792 critical and 10,561 high-severity findings.
• ChatGPT for Excel extension now available. Sidebar access to ChatGPT directly inside workbooks.
• Cursor expands its marketplace with more than 30 new plugins. Read, write and take actions across a wider range of developer tools and services.
• Notion launches Dashboards. A database view combining tables, boards, charts and timelines in one layout.
• Karpathy open-sources Autoresearch. An agent that runs ML training experiments autonomously on a single GPU, finding 20 real improvements and an 11 percent speedup after two days on 8xH100.
• NVIDIA releases Nemotron 3 Super, an open 120 billion parameter reasoning model for multi-agent AI. 12 billion active parameters via mixture of experts.
• Notion Custom Agents are free until 4 May, then move to a credit-based model. After 4 May, each agent run costs credits ($10 per 1,000 credits, roughly 30 to 60 runs). People are not happy about the pricing.
• Andrew Ng releases Context Hub. An open source CLI that gives coding agents access to current, versioned API documentation.
• Google open-sources Always On Memory Agent. Persistent memory for AI agents without a vector database.
• Hugging Face publishes a Synthetic Data Playbook. A practical guide for generating trillion-token training datasets at scale.
• AI assistants now account for 56 percent of global search volume. ChatGPT handles 89 percent of all AI search sessions.
• Meta delays its Avocado model after performance concerns. It does not match leading models from Google, OpenAI and Anthropic. Pushed to at least May.
• Anthropic launches the Claude Partner Network with $100 million in 2026 funding. Accenture, Deloitte, Cognizant and Infosys are the launch partners. Accenture is training 30,000 professionals on Claude, and Cognizant has opened access across its 350,000 workforce.
• Anthropic announces The Anthropic Institute, a new research arm led by co-founder Jack Clark. Interdisciplinary teams of ML engineers, economists and social scientists studying AI risks, labour market impact and how AI interacts with the legal system.
• OpenAI's robotics lead resigns over the Pentagon deal. Caitlin Kalinowski quit citing missing guardrails on surveillance and lethal autonomy. The first senior exit over the partnership.
• The US Senate approves ChatGPT, Gemini and Copilot for official staff use. Drafting, research and briefing preparation are now cleared.

How to build your first MCP App with Claude Code

MCP Apps let your tools render interactive dashboards, forms and visualisations directly inside Claude and ChatGPT. In this video we build a full MCP App from scratch using Claude Code and the official agent skill. The same approach works with any agent tool that supports skills, including VS Code Copilot, Goose, Cline, Gemini CLI, Codex, Cursor and Google Antigravity.

New to Claude Code? Start here. Then watch How to add interactive UIs in Claude and ChatGPT via MCP Apps.

Claude Code updates this week

Here are the highlights from the changelog.

• A one million token context window for Opus 4.6. Available by default on Max, Team and Enterprise plans. It doubles what was previously available and makes long codebases and large PRs significantly easier to work with.
• Tools can now ask you questions mid-task. Instead of guessing, connected tools can pause and ask for clarification through interactive dialogs before continuing.
• The /effort command. Set the model's reasoning level to low, medium or high directly from the prompt. Useful for quick lookups versus deep analysis.
• Faster startup on large codebases. New monorepo support lets agents load only the folders they need instead of pulling down the entire repository.
• The /btw command. Start side conversations while Claude is working. Ask a quick question or give additional context without interrupting the current task.
• Better memory and model control. New settings let you choose where Claude stores its memory files and connect to custom AI providers like AWS Bedrock and Google Vertex.

Claude Code Bootcamp, 24 March

Our first Claude Code Bootcamp is almost here, and we cannot wait. 24 March at Mesh Oslo. A full day of hands-on Claude Code, and no, you do not need to know how to code.

The group taking shape is a great mix: founders, tech leads, advisors and operational leaders from finance, maritime tech, media and consulting. Some are developers, most are not. That is exactly the kind of group this is built for.

If you are a project manager, consultant, analyst, lawyer, researcher, marketer or founder who wants to seriously level up your output, this is your day. Just bring a laptop and an open mind. A few spots left. Newsletter subscribers get 20 percent off with the code CCGRITAI03.

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 building. Join the GritAI community.

Best, Aleksander and Simon