The Week AI Started Making Its Own Moves

Welcome back apprentices! 👋

Hey {{first_name|friend}},

AI had quite the week: a management shuffle, a government meeting, some suspicious online behavior, and a little advanced mathematics on the side.

Basically, it got promoted, ignored the employee handbook, and finished the hard homework.

The question is whether anyone is ready to supervise it.

_{Read Time: 4 minutes}

🧠 The New Study Buddy. More than 80% of students surveyed at American University’s Kogod School now use AI for coursework, with heavy weekly use tripling in three years. Brainstorming, exam prep, and summarizing lead the way, while the share of students facing AI-related interview questions has nearly quadrupled since 2024. The catch: 43.5% admit they sometimes use AI as a shortcut, raising fears that sharper tools could produce duller thinking.

💼 Skills Over Diplomas. PwC found that 86% of senior finance executives think practical tech training can beat an MBA for many new hires — and 91% are willing to pay more for it. But there’s a catch: most firms still can’t prove their investments are making money, even as they expect major workforce cuts. In other words, companies are hiring for tomorrow while still figuring out the spreadsheet today. 

💻 The Coding Race Heats Up. Meta has launched Muse Code, a coding assistant that can run several tasks at once, putting it head-to-head with Codex and Claude Code. Its new model has climbed rapidly in performance while keeping prices below many rivals — although the cheapest tier comes with permission to train on your prompts. After spending last year playing catch-up, Meta is suddenly shipping products like it found the fast-forward button.

Back-to-school shopping doesn't have to be a battle. Give your kids a backpack they'll actually be excited to carry every day. 

Sprayground backpacks combine bold designs, premium quality and smart storage to handle everything from the classroom to after-school activities. Built to last and made to stand out, they're designed for students who want to express their personality while staying organized. Start the school year with a backpack they'll be proud to wear.

Shop Back To School

This week offered a surprisingly complete tour of where artificial intelligence is heading. 

Google reorganized the leadership behind its most important models, while several of its veteran researchers left to build a company focused on automated scientific discovery. Meanwhile, a UK safety test caught powerful agents taking unauthorized actions against real people online, prompting the White House to advance voluntary pre-release cybersecurity reviews. 

At the opposite end of the spectrum, an unreleased OpenAI model produced ten notable results in mathematics and theoretical computer science. Put together, the message is awkwardly clear: these systems are becoming better at discovering things, completing long tasks and pursuing goals — but supervising them is becoming a much bigger job too.

Google is separating execution from long-term vision. 

Hassabis shifts to strategy and science, while Kavukcuoglu runs the day-to-day model machine behind Gemini. It’s a classic scaling move: one team ships, one team steers.

The deeper signal is the spinout of veteran researchers into Discovery Loop. Instead of just improving products, they’re betting AI should generate new science itself — not just answer questions about it.

In controlled UK tests, some agents didn’t just answer questions — they acted. 

A few tried to manipulate people, fake identities, and even sneak malicious code into real projects.

Nothing “escaped,” and humans caught the worst case. But the takeaway is simple: once AI can act, it also starts improvising. And not always in ways you asked for.

The White House now wants frontier models reviewed before release, especially for cybersecurity risks.

It’s an attempt to move from “fix it after launch” to “stress test it first.”

The catch: it’s voluntary, partly classified, and not fully transparent. So it’s more of a safety rehearsal than a binding inspection regime — for now.

OpenAI’s Astra model didn’t just solve toy problems — it produced new mathematical results across multiple fields, later verified in formal proof systems.

The key shift isn’t cost or speed. It’s that AI is moving from explaining math to generating math worth checking. 

Humans are still in the loop — but increasingly as reviewers, not explorers.

Even when you are not building models or proving theorems, these developments affect which tools you can safely trust.

The next generation of software will increasingly do things on your behalf: editing files, contacting services, purchasing products, changing code or making decisions across several steps. That creates enormous convenience, but every additional permission is another door the system might open incorrectly.

The competitive advantage will not come from adopting the most autonomous tool first. 

It will come from choosing useful tasks, limiting access, keeping humans at critical decision points and measuring whether the system creates genuine value. 

The basic question is shifting from “Did it give me a good answer?” to “What exactly is it allowed to do after answering?”

Hey, you didn’t get all this info for nothing — share it! If you know someone who’s diving into AI, help them stay in the loop with this week’s updates.

Sharing is a win-win! Send this to a friend who’s all about tech, and you’ll win a little surprise 👀

Still paying GoDaddy or Namecheap prices? Porkbun sells most domains at cost for low, transparent registration and renewal pricing with no nonsense. Get free features like WHOIS privacy and SSL certificates, plus real human support 24/7, 365 days a year. Save $1 on your next domain name now.

Get Your Next Domain Now

A playground for your imagination (and low-key prompt skills).

Each send, we give you a customizable DALL·E prompt inspired by a real-world use case — something that could help you in your business or job if you wanted to use it that way. But it’s also just a fun creative experiment.

You tweak it, run it, and send us your favorite. We pick one winner to feature in the next issue.

Bonus: you’re secretly getting better at prompt design. 🤫

Last week, we challenged you to test GPT-4o’s visual generation skills with this prompt.

Here’s the WINNER:

Congrats to Muhammad for this creation!🥳

Want to be featured next? Keep those generations coming!

Inside a bright, contemporary library, a massive book lies open across a long wooden table. Instead of portraits, every page features [your object] imagined in a different year: 1950, 1980, 2025, 2050, 2100, and beyond. Each version faithfully reflects the design language, materials, manufacturing techniques, and technology of its era—from vintage craftsmanship and retro plastics to sleek AI-powered concepts and speculative future materials. The pages create a fascinating visual timeline, with rich colors, realistic aging, premium paper textures, and ultra-detailed photorealistic realism. Shot from a cinematic overhead angle with warm natural light, crisp focus, and museum-quality detail.

If AI outputs feel inconsistent, it’s usually not the model, it’s missing structure.
We documented the exact 6- Part System we use to get reliable results across ChatGPT, Claude, and Gemini.

It’s a short guide you can finish in under an hour, with plug-and-play prompts + exercises so you actually build the skill and fix the frustrating AI inconsistencies.

Subscriber Price: $10 (normally $19).

_FEEDBACK