How selling stink bombs helped me learn AI

A framework for learning AI as Cloud Security pros

Author

Nelson R
May 17, 2025

Howdy my friend!

In today’s piece:

  • How selling stink bombs helped me learn AI (a mental framework you can use too)

  • Orca enters the AI agent game by acquiring Opus Security

  • Google’s AlphaEvolve makes new mathematical discoveries

  • Other handpicked things I think you’ll like

In high school, I hatched a money-making scheme with two friends: selling stink bombs around the school.

It was the latest million-dollar idea in a line of failed ventures. My previous hustles included selling playing cards, virtual items in Diablo II, and a lawn care service called “The Leaf Brothers” (tagline: “Leaf it to us”).

None made me a millionaire. Or even a hundredaire. But stink bombs? “This is the way,” I thought.

Giddy with excitement, we bought a pack of 72 stink bombs from eBay.

Then we had to work out what to do next. We hadn’t planned that far ahead.

Step one was to test the product on unsuspecting friends to make sure it worked. One friend, Brian*, was top of our “they’d be fun to annoy” list, so he really copped it. He was not impressed.

Step two was marketing the product.

A word from our partner…

Start learning AI in 2025

Keeping up with AI is hard – we get it!

That’s why over 1M professionals read Superhuman AI to stay ahead.

  • Get daily AI news, tools, and tutorials

  • Learn new AI skills you can use at work in 3 mins a day

  • Become 10X more productive

Key Developments

Nelson’s summary: 

Leading Cloud-Native Application Protection Platform (CNAPP) Orca just acquired Opus Security, a cutting-edge startup that does agentic AI vulnerability management.

Why it matters:

This points to a major shift away from detection-focused tooling in the cloud security industry, and into a new era of auto-remediation via agentic AI.

In the words of Orca CEO and co-founder Gil Geron:

“With the acquisition of Opus, we’re moving decisively into the era of intelligent action, where cloud security problems are not just seen, but solved automatically.”

Nelson’s summary: 

DeepMind just announced AlphaEvolve, a powerful AI coding agent that discovers new algorithms to solve technical problems. This includes making more efficient data centers, chip designs and even AI training processes. It uses the creativity of the Gemini large language models to generate possible ideas, then automated evaluators to verify these ideas.

AlphaEvolve discovered a new algorithm for matrix multiplication that beat the previously known best algorithm.

When applied to over 50 open mathematical problems, AlphaEvolve rediscovered the state-of-the-art solution around 75% of the time. In 20% of cases, it improved on the previously best known solutions.

Why it matters:

This is a jaw dropping announcement. Google describes AlphaEvolve as a “general-purpose algorithm discovery and optimization” agent.

This is another way of saying: we’ve just invented a tool that discovers other tools.

A form of “automating automation”.

Almost all technological progress has been downstream of theoretical algorithm discoveries, including computers, modern communications infrastructure and transport systems.

So, finding a way to automate algorithm discovery is… huge. This system can be applied to “any problem whose solution can be described as an algorithm, and automatically verified”.

Other Handpicked Things

ZEST AI security: AI agent platform for remediating cloud security risks. Not sponsored & I haven’t tried it, but looks interesting. If you try it, I’d love to hear how you found it.

Windsurf’s SWE-1: Vibe-coding startup Windsurf launches SWE-1, it’s own in-house AI model designed for software engineering. Definitely progressed past the “ChatGPT wrapper” label that was thrown around in the early days.

Guidde: Cool tool to quickly create video documentation with AI.

Scrybe: Nifty tool to create LinkedIn posts rapidly, including finding viral posts for inspiration.

Continuous Thought Machines: a novel AI model that more closely resembles biological neural networks and might “think more like we do”. Another day, another funky AI architecture.

Claude’s 24k token system prompt leaked: Claude’s system prompt was published. It’s a big one: over 24,000 tokens of instructions.

Before you go, I’d love to know what you thought of today’s newsletter. That way I can improve the experience for you and minimise suckiness.

What'd you think of this email?

Login or Subscribe to participate in polls.

Take care.

Nelson, fellow AI enthusiast and cloud security dude.