ZD 25.43: The Week The Cloud Fell Down

The week we all learned it is always DNS.

Redundancy is Hard and Inconvenient

AWS suffered a catastrophic failure. It turned out the trusted platform for redundancy and resiliency had it’s own single point of failure. The old sysadmin in me was not surprised to learn it was DNS that caused the outage. There are a few things that are very, very hard to make redundant and the “where are you” system is one of them.

I am sure a small army of AWS engineers will come up with a plan to make this point redundant as well. There will surely be a next time, you cannot ever design this out of the platform. It is impossible to test for everything in a system of this scale.

To some extent it is on the builders to be prepared for this. Do not trap your users in their smart beds. As you think about the products you are building and the system design choices you are making, presuming that the chaos monkey will strike and that major pieces of infrastructure will die. Local failsafe and the ability to cold start are important. A tool needs to stand on it’s own in some way for your consumers.

The Distilled Spirit

Scandals and Bubbles

🎲 A Gambler’s Take (Silver Bulletin)
It isn’t shocking that there is a gambling problem in professional sports. Nate Silver, a professional gambler, does a great job covering all the angles here.

♻️ AI’s Circular Deals (Noah Smith)
Are AI companies making circular deals to hide losses? Or is this a case of major manufacturers funding an ecosystem not unlike the automotive giants of late?

App Routers, Scale and Timelines

↗️ Routing Towards AGI (The Diff)
OpenAI is building an intelligent switchboard to route your thought and ultimately buying activity through it’s engine. A machine that won’t connect one huge AI, but rather may small AIs. But what will that do for everyone’s margins?

🚨 Resiliency and Scale (Stratechery)
Notes on the AWS fiasco and other tightly-wound systems that are failing such as rare earth minerals and the information ecosystem.

📈 Still Early on the AI Timeline? (Rex Woodbury)
1995 was very early in internet years, are we in the same place with AI? The application layer is just forming. Change might be happening a bit faster this time but similar challenges remain.

Leading and Working

🧑‍🎨 The Genius of Amateur (The Works in Progress Newsletter)
One needs amateurs and outsiders to create conditions and ideas that lead to true advances and progress. You need to think outside of the box.

🍎 What Steve Jobs Got Right and Wrong (Guy Kawasaki)
Steve Jobs is both an inspirational and cautionary example for leaders.

🥳 How To Reintegrate Your Individuality (Still Wandering)
Five ways to figure out what you enjoy to help you find the next thing. This writer thought #2 is especially profound.

🦸 ADHD is Just a Personality Trait (Michael Woudenberg)
ADHD is gets described as a disorder or a superpower depending on setting. Is it just simply a personality trait that gets overanalyzed?

AI: Philosophical

🎨 Seeing Creativity Like a Language Model (Every)
Dan Shipper’s work journey was reshaped by AI. Once you realize your art is more than the output — the pixels or ink on a page — but something more you can understand how the tool can be a boon to your creativity in the allocation economy.

📖 You Cannot Outsource Understanding (A Software Enchiridion)
Your developers perform a technical function, but they are also people who understand your product and your process. Do not repeat the mistakes of offshoring with AI.

⛏️ Thoughts on the AI Buildout (Dwarkesh Patel)
Wrap your head around energy and AI capex with this detailed post. It turns out natural gas is much faster and really not that much more expensive compared to the other parts.

AI: Practical

📺 Automated Youtube Tracking (The Circuit)
There is more YouTube content to track than hours in the day. Using a little Make automation and AI, you can build a theme tracker that will drop a daily digest into a Notion database without much time nor trouble.

🕵️ Build Your First AI Agent (AI Disruptor)
Step-by-step instructions for creating a workflow with the new OpenAI Agent Builder tool.

💵 Research Stocks Like a Pro (Jimmy's Journal)
How to use research tools to simulate the scuttlebutt method of researching a stock.

💡 Saving Time With Claude Skills (AI Adopters Club)
Claude skills are a new way you can interface with tools and also encapsulate workflows. They are a bit like more advanced Custom GPTs if you are used to OpenAI’s implementation. Here is a good example of how to get going with one.

✍️ The Definitive Guide to Claude Skills (Limited Edition Jonathan)
Everything you need to know on a nuts and bolts level about implementing Claude skills. Really good when your run into trouble or want to ship the skill across a workplace.

🙈 Why AI Sometimes Forgets (ByteByteGo)
An explainer on how AI loses it’s context and memory at times and how RAG can help.

AI: Deeply Technical

💎 30 Tips for Using Gemini CLI (Elevate)
Google has published a powerful command line interface for Gemini. It can do much more than code — it can do things like clean up files and interact with google documents and sheets.

📚 Production RAG Hints (Abdellatif Abdelfattah)
Lessons learned from 8 months in the RAG trenches processing upwards of 5 million documents.

🔗 Contextual Power of the Hyperlink (Michael Bleigh)
Links provide a very powerful scaffolding for context management in your LLM applications.

📘 Build Your Own DB (Not a Number)
Ever wondered how a simple database works? Here is an interesting example.

Interesting Things

😎 What Recording Light? (404 Media)
Tired of the pesky recording light on your Meta Ray Ban smart glasses? Disable it for $60!

💄 Notes from Korea (ChinaTalk)
Two correspondent’s notes about life in South Korea from a recent trip.

🏃‍♂️ The New Run Club (Daniel-Yaw Miller)
Calisthenics crews are the new run clubs. Everyone is doing it.

The Look

Amazon is building a robot army.

Did you enjoy reading this post? Hit the ♥ button above or below because it helps more people discover great Substacks like this one and it helps train your algorithm to get you more posts you like. Please share here or in your networks to help us grow!