THE AI UPSKILL

A weekly digest for Beginner Builders

⚡ TLDR

Anthropic just became the second most valuable private company in history, and a coding tool most people have never heard of is the reason why.

Meanwhile, the developer behind OpenClaw, the AI agent that took over the internet last month, turned down millions to join OpenAI instead. And while everyone was distracted, Google quietly posted AI test scores so dominant they made the competition look like it was running in a different race.

Also in this week's newsletter, we're building an AI email classifier that reads and sorts your inbox automatically and sharing our top AI video picks from the week.

ARTIFICIAL INTELLIGENCE
THIS WEEK WE'RE WATCHING

How a college student made $500k with AI : Nate Herk has another banger this week, interviewing a college student who made bank with grit, cold out reach and a willingness to the odd project for free. It’s inspiring, and a reminder that there are no excuses. The potential to make serious money with AI is at your fingertips. Check it out here.

Sabrina Ramonov 1-hour Claude Course for beginners: Yes, she plugs her product Blotato, but even still, this video is full of gems and insights for new Claude Code users. By the end of the video, you’ll have learned how to run your entire social media operation on autopilot. Watch it here.

THE BUILDERS TIP
AI PLAYGROUND: POWERPOINT PREZ WITH CLAUDE

If you're on a paid Claude plan and haven't discovered this yet, you're in for a treat. Most people use Claude like a smarter search engine: ask a question, get text back. But buried in your settings is a feature called Skills, and once you switch it on, Claude stops just describing things and starts making them. Actual files that are ready to download and use. 

While PowerPoint is the most impressive, it works for Word documents, Excel spreadsheets and PDFs too. The kind of thing that used to take an hour now takes one prompt.

Here's how to turn it on

  1. Go to Settings → Capabilities in claude.ai

  2. Enable Code execution and file creation

  3. Toggle on the Skills you want (PowerPoint, Word, Excel, PDF are all there)

That's it. You're done.

Need inspiration, try these prompts:  

  • "Create a 5-slide PowerPoint presentation introducing AI literacy for marketers. Include: a title slide, what AI literacy means, why it matters, three practical skills to develop, and a summary slide.”

  • "Make a one-page PDF summary of [paste any article or text]”

  • "Build a simple Excel tracker for my weekly content calendar with columns for platform, topic, status and publish date"

  • "Make a 3-slide deck I can use to pitch AI tools to a sceptical client"

LET”S BUILD
SIMPLE BUILD: AI NEWS AGGREGATOR

Your inbox is chaos. Gmail's filters can sort by sender, but they can't read an email and decide what it's about. This build can.

It uses an LLM to read every incoming email, classify it into one of five categories, and auto-label it in Gmail. No rules to maintain. The AI figures it out.

Categories: Newsletters · Work · Transactions · Promotions · Others

QUICK START:

  1. Download the workflow JSON here. 

  2. Paste it into your N8N instance

  3. Add your Gmail OAuth and OpenAI credentials

  4. Create five labels in Gmail (Newsletters, Work, Transactions, Promotions, Others)

  5. Replace the placeholder label IDs in each Gmail node with your actual label IDs

To find your Gmail label IDs: In N8N, add a Gmail node, set the operation to "Get Many Labels," run it once, and copy the IDs from the output.

Something not working? Copy the node input and output, then paste it into Claude for troubleshooting.

MAKE IT YOURS:

  • Change the categories — Want "Client Emails" or "Job Applications" instead? Open the Classify Email node and rewrite the descriptions. The more specific you are, the better it classifies.

  • Tweak accuracy — Run it for a day. If newsletters keep landing in Promotions, paste a misclassified example into your LLM and say: "This was classified as Promotions but should be Newsletters. Update the prompt to handle this."

  • Swap the model — We use gpt-4o-mini for cost (~$0.01 per 100 emails). Want better accuracy? Switch to gpt-4o or Claude Sonnet.

Add actions beyond labeling — Auto-archive promotions, forward work emails to Slack, save newsletter links to Notion. The classifier is just the starting point.

IN OTHER NEWS
THE BIGGER PICTURE

Anthropic Is Now Worth More Than Goldman Sachs. It Didn't Exist Four Years Ago.

Last week, Anthropic (the company behind Claude) raised $30 billion from investors at a valuation of $380 billion. It is the second-largest private funding deal in tech history, making Anthropic more valuable than Nike or Goldman Sachs. It’s even more staggering when you realise that Anthropic was only founded in 2021. 

Eight of the ten largest companies in the world are now paying customers. Two years ago, a dozen companies spent over $1 million a year with Anthropic. That number is now 500.

Our take: The old software tools that companies have paid for over the past 20 years are being replaced by AI that can do the same jobs faster and cheaper. Every dollar going to Anthropic is a dollar not going somewhere else. The wave is coming. Stay ahead of it. 

Google Builds Text-to-Game AI 

OpenClaw Went Viral. Then OpenAI Showed Up

For a few weeks earlier this year, one app was all anyone in Silicon Valley could talk about. Not something built by Google or Meta or a well-funded startup. Just OpenClaw, a side project by Austrian developer Peter Steinberger, quietly built in his spare time.

The idea was deceptively simple: instead of asking AI to do things every time, OpenClaw runs in the background and acts for you. Booking flights, managing your inbox, scheduling your week. You set the goal, it gets on with it. Hundreds of thousands downloaded it almost overnight.

This week, OpenAI hired Steinberger to help build "the next generation of personal agents." OpenClaw will move into an independent foundation and stay free and open source.

Our take: The AI industry is in a full sprint toward tools that don't just answer questions but take action on your behalf, and the big labs are willing to pay whatever it takes to get the right people. Steinberger didn't have a team or venture capital. He had a sharp idea, the skills to build it, and he shipped it. The opportunity right now isn't just in using AI. It's in building with it. The gap between "person with an idea" and "person who can actually make it" has never been smaller.

🔥 THE WATER COOLER: 

The AI Leaderboard Just Changed. Google Is Now Pulling Away.

Every few months, the big AI companies publish test scores to prove their models are getting smarter. This week, Google's Gemini sat the tests - and wiped the floor. 

On one test measuring how well an AI handles problems it has never seen before, Gemini scored 84.6%. Claude scored 68.8%, while OpenAI’s GPT scored 52.9%. A 16-point gap in AI benchmarks is the difference between good and generational.

But perhaps a bigger story is something Google quietly launched called Aletheia. It's an AI that can tackle unsolved maths problems, check its own working, and write up the findings as a research paper. Without a human touching it at any point. The kind of work that could take a PhD mathematician months.

Our take: You've probably heard people say AI will eventually do the work of scientists and researchers. Well, that time is approaching faster than we think.

(Read more: Google Deep Mind)

That's it for this week.

Building something cool with AI? Reply to this email—I feature reader projects. 💬

P.S. If this was useful, forward it to someone upskilling in AI.

Keep Reading