Welcome, humans.

What happens when you attend a dinner party hosted by Dwarkesh Patel, one of AI's most connected podcasters? You overhear the conversations that AI insiders have when they think no one's listening.

A recent party recap dropped some wild insights. Turns out, AI could have happened decades earlier, but a famous 1992 research paper convinced everyone it wouldn't work—based on testing neural networks with only four brain cells. Four! That's like concluding cars don't work after testing a bicycle.

Someone also did the math on what it would cost if ChatGPT could actually update its brain based on your conversations. Not just reference old chats, but permanently get smarter about you specifically. Price tag: around $20-100 per month. Not bad for a personal AI that grows with you.

(P.S: Dwarkesh just released an excellent post about why AI will need continual learning like mentioned above to actually do human-like work).

Perhaps most tellingly, one guest asked Claude about a technical AI concept, got a confidently wrong answer, and realized something scary. As these AI systems get better at sounding convincing, how will we know when they're just making stuff up?

The dinner party conclusion? “LLMs will only get more convincing from now on.” Great for productivity, concerning for truth.

Here’s what you need to know about AI today:

• We break down Microsoft’s 1 hour agent course.

• Claude now has a Deep Research mode.

• Lawmakers from all 50 states oppose new AI “anti-regulation.”

• Godfather of AI launches $30M nonprofit to improve AI.

Microsoft Just Dropped an Hour-Long Free Course to Turn You Into an AI Agent Builder.

Tired of hearing about “AI agents” all the time? Still not sure what they are, or how to build one? Microsoft’s got you covered. They recently released a free, one-hour “AI Agents for Beginners” course, broken into ten 5-minute lessons.

So, what’s an agent? Microsoft says to think of it as a smart AI with:

• A Brain (LLM): For understanding, planning, and reasoning.

• Memory: Short-term (current chat) and long-term (your preferences).

• Tools: Access to external functions or data (like flight booking or weather apps).

The course guides you through building “agents” using frameworks like Semantic Kernel (for production agents) and Autogen (for multi-agent experiments), often with free models via GitHub.

Here’s a rapid-fire look at what you'll learn:

1. Designing Good Agents: Learn how to make your AI helpers easy to find, clear about what they can (and can't) do, and smart enough to get better by remembering past chats.

2. Giving Agents Tools (Tool Use): Discover how to let your agents actually do things by giving them access to real-world functions, like booking a flight through an actual booking system.

3. Smarter Information Lookup (Agentic RAG): See how agents can understand complicated questions, find information from different places (like your company documents and a live online service), and then put it all together for a complete answer.

4. Breaking Down Big Tasks (Planning): Find out how agents can take a big request, like “plan my Paris trip,” and chop it into smaller, manageable steps like booking the flight, then the hotel.

5. AI Teamwork (Multi-Agent Systems): Learn how to get multiple AI agents to work together as a team, where each one might have a special skill, like one agent writing a draft and another agent reviewing it.

6. Learning As They Go (Metacognition): Understand how to design agents that can “think about their own thinking,” noticing your preferences over time and using that to give you better suggestions in the future.

7. Getting Ready for Real Users (Putting it in Production): Learn the important steps for making sure your agent works well when real people start using it, including how to check its performance and plan for when things might go wrong (like an online service it uses temporarily breaking).

Why it matters: AI agents *should be able to automate much more complex tasks than current chatbots. By making this knowledge accessible, Microsoft is lowering the barrier to entry for developers and even curious non-coders (like you, perhaps??) to start building more sophisticated AI applications (don’t worry, if this over your head, we share some more user-friendly agent tools in the TTT below!)

Check out the full AI Agents for Beginners course on GitHub here. It includes all the written lessons, code samples (Jupyter Notebooks), and the video walkthroughs.

FROM OUR PARTNERS

StoryAI by Fullstory 

Meet StoryAI: your shortcut to smarter decision-making.

Powered by conversational AI agents, teams can analyze multiple user sessions, get instant answers to key questions, and predict behavior with confidence.

Turn insights into action and data into outcomes—faster than ever. The future of behavioral data starts right now—like, as soon as you click the button below. Just saying.

Get a Demo Today

Prompt Tip of the Day

Something we’ve been doing lately when we need to summarize something is asking our AI to summarize “with 100% fidelity to the original.” This phrasing helps the AI attempt to maintain the original meaning of the text even when condensing it.

We saw some similar prompt advice in this thread, which has some other good ideas too, such as this section of the prompt:

The rest of the prompt might not be necessary depending on what you need, but try the above part out and see if it improves your summarized / condensed outputs!

Treats To Try.

*Asterisk = from our partners. Advertise in The Neuron here.

1. *Incogni removes your personal data from the open internet so scammers and identity thieves can’t access it. Protect yourself online with Incogni—get 55% off with code NEURON.

2. n8n is a way to create agentic workflows (if not full-on agents) that’s open-source and free for you to self-host.

3. If you prefer somebody else hosting your agents and don’t mind paying, check out Lindy or CrewAI for enterprise.

4. Zapier and Make are two other tools that aim to make it easier to connect multiple tools and services together for an AI to use in sequence.

5. ShumerPrompt is a nice little prompt library from Matt Shumer with a pretty slick UI; it’s sorta like a grown-up version of our own Prompt Library experiment.

6. Claude now has a Research mode (similar to ChatGPT’s Deep Research), which are both basically agents that research the web for you.

7. NotebookLM now lets you share your research notebooks publicly with anyone so others can access your study guides or listen to AI-generated podcast summaries of your content.

See our top 51 AI Tools for Business here!

Around the Horn.

• Here’s a great video about corporate “AI washing” and how companies are using the keyword “AI” to justify massive investments or headcount cuts; the funniest part is that a company called BuilderAI was literally just using humans in India posing as AI (hence the joke AI = Actually Indians).

• Over 260 US state legislators from all 50 states opposed a federal proposal for a 10-year ban on state AI regulation.

• Anthropic apparently has restricted vibe-coding tool Windsurf's access to Claude models.

• The Trump administration reorganized the U.S. AI Safety Institute into the Center for AI Standards and Innovation.

• OpenAI launched internet access for its Codex software engineering agent for ChatGPT Plus subscribers.

• Yoshuo Bengio (a so-called godfather of AI) raised $30M for LawZero, a non-profit to make AI “less-human-like.”

• Claude has a blog where “Claude Explains” complicated code topics to you.

Under the Hood

1. AgenticSeek is a fully open AI agent that browses the web, codes, and plans tasks locally on your device (try this instead of paying cloud subscriptions).

2. Here’s how to master Claude Code in 30 minutes, and everything to know about OpenAI Codex in 5 minutes as well as a slightly longer, 19-min in-depth tutorial from Helena Liu.

3. If you’re going to get real serious about building AI agents, then you’re going to need to learn about building MCP servers—and this video explains that REAL well.

4. Also, Claude released some new agent-building capabilities in the API, as did Mistral.

5. Landing AI from Andrew Ng has a new document extractor agent—try it here.

6. GitHub Agent Mode completes entire coding projects for you by automatically editing multiple files, running terminal commands, and fixing errors as they happen (read more).

7. Are you a software engineer who wants to become an AI engineer? Then this podcast interview w/ Janvi Kalra is for you!

8. If you’re an AI engineer and “want to be 5 years ahead” in AI land, you could read all of DeepSeek’s papers.

9. This is another good X thread of “the top 50” AI papers since 2017.  

A Cat's Commentary.