Hey, it’s Andreas.
I’m racing toward two book deadlines, which explains the very intermittent schedule of Human in the Loop over the past few weeks.

I’m especially excited about publishing what may be one of the first major practical books on Claude Code with a major publisher. At least from what I’ve seen, it could be the first book that goes this deep on the topic.

I’ve been living in Claude Code from morning to night, and I know many people who are doing the same. Over the past months, I’ve collected a huge amount of practical knowledge, workflows, patterns, and lessons, and turned it into a book. I think it will be incredibly useful.

My other book is about building a career in generative AI. It is co-authored with colleague’s from Google, Microsoft, and Liquid AI, and it is already up to pre-order. You can have a look here.

My goal with writing is simple: whichever side of the AI debate you lean toward, I want to help people understand what is happening and start using AI tools now, seriously.

That’s also why I’m exploring a multi-week AI Agent Mastermind cohort for business professionals. Not a basic overview course, but a a weekly live format where we build from day one. The goal is to help professionals like you to become more productive, automate meaningful parts of your work, and potentially build AI-powered products you can generate income with.

I want to keep it intentionally limited and highly selective - for professionals who are done watching AI happen from the sidelines and are ready to build. If this sounds interesting, let me know.

—
In today’s issue:

Let’s get into it.

🌀Anthropic overtakes OpenAI in valuation → Anthropic raised $65B at a $965B post-money valuation, putting it ahead of OpenAI’s last reported $852B.

🌀Nvidia unveils RTX Spark for personal AI agents → A new superchip built to bring up to 1 petaflop of AI performance to Windows laptops with all-day battery life.

🌀Apple rebuilds Siri around Google Gemini → First leak on Apple’s long-delayed AI Siri: dynamic Island interface, AI search, richer cards, a ChatGPT-style Siri app, and support for external models and third-party agents.

🌀Microsoft reportedly plans a Copilot super app → GitHub Copilot, chat, Cowork, and Autopilot could merge into one AI work hub.

🌀Perplexity open-sources Bumblebee → A local safety scanner for developer machines that checks risky packages, browser/editor extensions, and AI agent configs - without executing them.

🌀Inherent Labs launches Faraday for self-improving AI science → Ex-DeepMind team raises $50M to build a science platform where researchers work alongside agents that don’t just answer questions - they help decide which questions are worth pursuing.

📘 OpenAI Academy adds new coding and agent tutorials → OpenAI Academy is becoming a solid learning hub for builders, with new resources on Codex, agentic workflows, MCP, and production-ready agent architecture.

📘 Anthropic shares prompting playbook for agentic systems → A practical guide on applying core prompting principles to agents that plan, act, and adapt.

📘 Forward Deployed Engineers become AI’s hottest new role → Frontier labs are realizing the bottleneck is no longer the model - it’s making AI survive inside messy enterprise systems. Forward Deployed Engineers embed with customers, write production code, build evals, connect legacy systems, and own the path from pilot to production.

📘 Google launched 5-Day AI Agents Vibe Coding Course → Free intensive on building agents with tools, memory, security, and production deployment. Includes daily labs, livestreams, and a capstone where you build an autonomous agent in a farming simulation.

🔹 Jensen Huang says agents are the next compute platform → In his COMPUTEX 2026 keynote, Jensen Huang framed the next shift: AI agents are becoming the biggest consumers of compute. Worth to watch the full 2 hours.

🔹 Peter Steinberger shares an auto-review skill for agents → A practical agent workflow, which helps you to automate and review your setup.

🔹 Cursor on the new habits of AI-native developers → Cursor’s Developer Habits Report shows AI coding productivity is compounding, but unevenly. Weekly code output per dev more than doubled from 3.6K to 8.6K lines in 18 months, while agents are doing more end-to-end work and AI-authored commits are rising fast. Most interesting: the top 1% now produce 46x more code than the median user.

🔹 Thorsten Ball on “Software After Software” → Great read from Thorsten Ball, who runs Amp. His core point: software is moving from human-written code to agent-delegated work. Code gets cheaper, judgment gets more valuable, and the winning teams will reorganize around models instead of forcing agents into old processes.

🔹 OpenCode creator Dax talks in the pragmatic engineer → A useful counterweight to the AI productivity hype. Dax argues the gains are real for some workflows, but much less obvious across teams, systems, and messy production work.

🔹 Princeton researchers on testing long-horizon AI agents → New research shows that long, real-world tasks reveal agent capabilities that standard benchmarks often miss.

I am an optimist about AI. I think this is the most exciting time in human history. So I am probably the last person you would expect to sit through Magnifica Humanitas, a 42,300-word papal warning about technology, written by the Pope. But I ended up reading every single word, because Leo XIV made a point in his first encyclical worth taking seriously, and it turns out to be the point my optimism depends on.

Consider the strangeness of it: an institution that thinks in centuries has decided to comment on an industry that thinks in quarters. The Pope presented Magnifica Humanitas himself, on a stage he shared with a co-founder of Anthropic. That pairing alone tells you how unusual this moment is.

His core worry is not that AI is evil. It is that a moral AI counts for nothing "if that morality is determined by a few." The drivers of the technology, he points out, are private, transnational companies that already outstrip the capacity of many governments, and he wants it "disarmed" and made human-friendly before it turns people into cogs in an efficiency machine. You do not have to share his theology to notice that this is a clearer description of the power problem than most of what the industry says about itself.

Because the question is not whether AI will be powerful. It already is.

The question is: who gets to shape it, who gets protected from it, and what kind of human future we are actually optimizing for.

That is why one passage stayed with me. It is the reason I am writing about this in the first place:

Strip the theology and what remains is an argument about agency. The future is not something AI does to you. It is something you choose to build, and the choice is being made right now, by whoever shows up to make it. That is where the optimist and the cautious Pope stop being opposites. I think the upside is underrated; he thinks the risks are underpriced. We are pointing at the same mechanism: the good version of this future is not the default. It does not arrive on its own. It arrives if people build toward it on purpose.

If you are a founder, an operator, a consultant, or running a product or enterprise team, this is not a philosophy-seminar question. It is an operating stance, and it comes down to four things.

Do not outsource your opinion to the labs. They are brilliant and they are also, in the words of Anthropic's Chris Olah on that same stage, operating "inside incentives that can conflict with doing the right thing." Your read on what AI should do inside your organization is not theirs to make for you.

Do not become a passive critic either. The loudest move in the room is to call for a pause, a moratorium, a deceleration. It feels like conscience and functions like an exit. You collect the social reward for the brave-sounding objection and change nothing about the trajectory, because the funding, the incentives, and the geopolitics are already locked in. You cannot steer a thing you refuse to touch.

Pick your construction site. You will not influence all of AI. You can influence the part you actually stand on: a workflow, a product surface, a policy, a team's skills. Choose it deliberately instead of letting it be chosen for you.

Then build there. Build the workflows, the guardrails, the products that make the good outcome more likely on your patch of ground. As the encyclical puts it, technology takes on the characteristics of those who devise, finance, regulate, and use it. That is not a warning to retreat. It is a map of where your influence actually sits.

There is plenty in the 42,000-word letter I disagree with, and a fair amount I am not equipped to judge. I am not religious, not even Catholic, and I am not converting. But on the question under all of it, whether the serious people should build the future or narrate it from a safe distance, the Pope and I land in the same place. I will keep my optimism, because I think it is earned. It just comes with a condition attached, and the condition is the entire job: the most exciting time in human history only becomes the best one if enough of us treat it as something to build rather than something to wait for.

The full encyclical, "Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence," is published on the Vatican's website.