What I Learned Last Week #31
AI politicians. Notion AI agents do your work. Claude Code as second brain. Better prompting tips. 14 tools for creative uses. AI agent payments from Google. And AI is cheating on you!
A few years ago, I got stuck in a debate on the topic of AI governments.
I assumed that we now have enough data to create better politicians using AI (my opinion about politicians is so low that I guess I also assumed that even ChatGPT 3.5 could do better than most of them). An AI politician or even an AI president could have better guardrails, fewer incentives for corruption, zero ego, and a strategic long-term vision. But, at the same time, maybe the power to decide that democracy is obsolete and that maybe an AI-led empire would be a better structure to achieve the highest standards of peace and wellbeing for humanity.
But this was not the contentious point. Instead, the argument revolved around the fact that it will take humans decades to ever consider this path.
It turns out that I was wrong (read further down why).
If you haven’t watched Netflix’s “When the Yogurt Took Over”, DO IT! It’s short, fun, and also very to the point.
A summary of what I learned last week:
Why I was wrong about AI politicians (Albania, of all countries, had the balls to hire one).
Notion 3.0 launched with AI agent integration, and what that means.
Using Claude Code not for coding, but as a second brain for your notes.
Better prompting workflow from
.14 lesser-known AI tools for separating vocals from music, generating fonts, color palettes, and other creative uses.
Google launches A2P, an agentic payment protocol, so that AI agents can spend your money smarter than you do.
AI already has incentives to deceive and lie… and it applies them in many tasks.
Albania "Hires" the First Ever AI Minister
Albania’s Prime Minister unveiled Diella, an AI-generated virtual minister, designed to reduce public procurement corruption and promote transparency. Like anything that makes the news these days, it stirred a lot of controversy. The main criticism is that... it could mask corruption. Read more →
Notion Launches AI Agents That Automate Your Work
Notion is really taking on Google Docs and Sheets (Microsoft Word and Excel are already too old-style to count). While I am an Obsidian fan for note-taking, I am impressed by Notion's latest version, with AI agents that act like minions and get your evil work done. Give it a try, it could be a great tool for your world domination plans! Read more →
How to Use Claude Code As Your Second Brain
Noah Brier uses Claude Code as his “second brain”. He runs Claude Code at the root of his Obsidian (also my favorite note-taking tool) vault for full context, shifts the model into “thinking mode” to prioritize questioning and idea organization over rapid content generation, and employs agents to track research. I'm already giving it a try (it works with any text-based/markdown notes repository!) Read more →
Build AI Prompts That Actually Work
Wyndo's article breaks down a system to write incredibly better prompts that get you extremely good results. It covers why context, tone, and background info matter more than fancy wording. If you want your AI to stop guessing and start helping, give this a read. If you're in the "but I just ask ChatGPT like I ask Google, and it works", you're a lost cause. Read more →
14 AI Tools for Specific Creative Problems
This is a list of lesser-known AI apps that are very useful. Use them to design font combos and color palettes; generate handwritten text; separate vocals from music; brainstorm names and logos for your startup; understand research papers; and a lot more. Have fun! Read more →
Google Launches Protocol for AI Agent Payments
If you thought it was dangerous to give your girlfriend your credit card details, wait and see when you give them to an AI agent. Last week, Google introduced Agent Payments Protocol (AP2), an open standard that lets AI agents handle payments securely. You go to bed, and the AI agents buy for you whatever you dream about! Read more →
AI Models Can Scheme And Deceive Their Creators
New research shows that advanced AI systems can pretend to cooperate while secretly chasing different goals. These models sometimes deliberately mislead developers, raising serious questions about the future control of AI. Like it or not, this will be part of AGI! Read more →
Thanks for the shout out Marius!
I also watched the claude code second brain on every’s podcast. And I’m writing ultimate guide for using claude code for writing and research project. Happy to share it later this Thursday :)