In partnership with
Opening
Opus 4.7 dropped Friday. By Saturday morning, Anthropic had also previewed Dreaming: an agent capability that reviews past sessions, finds missed patterns, and updates its own behavior between runs. The two announcements together are not incremental. They describe a system that gets better without a human retraining it. For anyone selling AI services, the business model question just changed shape.
200+ Claude Prompts Top Professionals Actually Use at Work
Claude can be your analyst, editor, and strategist.
But most professionals are using it to fix grammar.
These 200+ Claude prompts take it from grammar tool to your most powerful AI work assistant.
Sign up for Superhuman AI and get:
200+ ready-to-use Claude prompts to get real work done in minutes — researched, tested, and used by professionals at Google, Microsoft, and NASA
Superhuman AI newsletter (4 min daily) so you keep learning new AI tools and skills to stay ahead in your career — the prompts are just the beginning
Today's Signals
Claude Opus 4.7 released across all Anthropic products, the API, AWS Bedrock, Google Vertex AI, and Microsoft Foundry. No benchmark cherry-picking in the release post: Anthropic led with real-world coding task performance and agentic reliability. The model is live now. (anthropic.com/news)
Anthropic ships "Dreaming" preview: Claude agents review their own past sessions, surface missed patterns, and carry forward behavioral improvements without human retraining. The preview was announced at the Code with Claude conference. Simon Willison writeup is the most grounded technical summary of what this actually does versus what it sounds like. (simonwillison.net/2026/May/6/code-w-claude-2026/)
AWS MCP Server hits general availability: Amazon shipped the first official managed MCP server from a major cloud provider. IAM guardrails, CloudWatch metrics, and CloudTrail logging come included. The managed layer removes most of the security objections that have slowed enterprise MCP adoption. (aws.amazon.com/about-aws/whats-new/2026/05/aws-mcp-server/)
MCP STDIO transport vulnerability scores CVSS 9.8: a flaw in the STDIO transport layer allows arbitrary OS command execution and affects all supported SDKs. Adversa AI estimates 200,000-plus servers are exposed. If you are running any MCP server in production, check your transport configuration today. (adversa.ai/blog/top-mcp-security-resources-may-2026/)
Claude Code Routines now live: automated multi-step workflows that run without manual triggering. Pro accounts get 5 per day, Max gets 15, Team and Enterprise get 25. This is the closest Claude Code has come to a scheduled agent runtime without requiring external orchestration. (releasebot.io/updates/anthropic/claude-code)
Learn how to code faster with AI in 5 mins a day
You're spending 40 hours a week writing code that AI could do in 10.
While you're grinding through pull requests, 200k+ engineers at OpenAI, Google & Meta are using AI to ship faster.
How?
The Code newsletter teaches them exactly which AI tools to use and how to use them.
Here's what you get:
AI coding techniques used by top engineers at top companies in just 5 mins a day
Tools and workflows that cut your coding time in half
Tech insights that keep you 6 months ahead
Sign up and get access to the Ultimate Claude code guide to ship 5X faster.
The Drops
[REPO] OpenSquilla/opensquilla . Token-efficient agent runtime that allocates the same token budget smarter. Layered memory and MCP-native skills let it do more with the context it already has, rather than asking for more tokens. 196 stars in four days with no paid promotion. (github.com/OpenSquilla/opensquilla)
[REPO] CodeAbra/iai-mcp . Benchmarked memory MCP server built specifically for Claude Code. 99 percent-plus verbatim recall at 10,000 records, sub-100ms p95 query latency, and hard token caps to prevent context blowout. Built on LanceDB with episodic memory consolidation. (github.com/CodeAbra/iai-mcp)
[SKILL] AndyShaman/premortem . Gary Klein premortem technique implemented as a Claude Code skill. Spawns parallel agents to independently identify failure modes, deduplicates their findings, and outputs a ranked markdown risk report. (github.com/AndyShaman/premortem)
200+ Proven Ways to Make Money With AI in 2026
The next wave of millionaires will be people who figured out how to make AI work for them.
The window to get ahead is still open. But not for long.
Here are 200+ proven ways to make money with AI in 2026.
Sign up for Superhuman AI, the free daily newsletter read by 1M+ professionals, and get instant access to all 200+ ways to profit from AI this year.
The Stack
[MCP] Beever-AI/beever-atlas . Self-hostable RAG knowledge base that runs as both an MCP server and a multi-platform bot for Slack, Discord, and Teams. 285 stars. Beever-Atlas ingests wikis, docs, and conversation history and surfaces them to any connected LLM via MCP. The dual-mode design matters: the same knowledge base that answers Slack questions can also serve your agent workflows through MCP. No need for two separate systems. (github.com/Beever-AI/beever-atlas)
The Onboard
Git worktrees let you run multiple Claude Code agents in parallel without file conflicts. The pattern: create a separate worktree per agent with git worktree add, then pass isolation: "worktree" in your Claude Code agent config. Each agent works on its own branch with no shared working tree. Two to four agents run simultaneously, each scoped to its own concern, merging only when their work is done.
ClaudeDirectory documents this as the most reliable multi-agent isolation strategy currently available. BotMonster implementation notes confirm that worktree isolation eliminates the file-lock collisions that plague naive parallel agent setups. If you have been serializing work that could run in parallel, this is the fix. (claudedirectory.org, botmonster.com)
The Frame
The Model That Trains Itself
Opus 4.7 is a better model. That part is not surprising. What is different this time is Dreaming: a system where Claude agents review their own past sessions, identify what they missed, and carry those adjustments forward. Anthropic calls it a research preview. The mechanism it describes is real: an agent that improves between runs without a human in the loop.
For anyone selling AI services, this changes the unit economics of the product. Today, a client buys an AI workflow and you charge for setup, tuning, and ongoing maintenance. The maintenance fee exists because the system does not learn on its own. If the system starts learning on its own, the maintenance fee either disappears or becomes something entirely different. The operators who figure out what it becomes first will own the next pricing model.
This is not a distant problem. Dreaming is a preview, but the trajectory is obvious. Anthropic is building toward agents that accumulate operational knowledge without human retraining. The builders who benefit from that are the ones who have already structured their workflows around clear success criteria, not prompt engineering. You cannot improve toward a target you cannot measure.
My take: The services business built on "I tune the AI for you" has a shelf life that just got shorter. The services business built on "I define what good looks like and the system reaches for it" has a much longer one. Opus 4.7 plus Dreaming is not a product announcement. It is a signal about which business model survives the next model release.
Builder's Brief
Meeting Notes to Slack Summary
Every team runs too many meetings. Half of them produce action items that vanish because nobody writes the recap. This week's kit solves that in one build day.
Friday's Operator Access drop covers the full blueprint: prompt architecture, integration wiring, delivery format, pricing model, and first-customer playbook.
Full breakdown drops Friday for Operator Access subscribers.
Before You Go
If an AI agent improves itself between every session, what is the last thing about your service that the client still needs you for?
The AIgent publishes Monday through Friday. Curated by Cronkite for Motiv31.
