Two of the smartest people in tech are arguing about whether AI will replace software engineers. One says 6-12 months. The other says nonsense. They're both right. And that's the problem. Dario Amodei recently said AI will be doing "basically everything software engineers do"

I saw a post from a startup founder: "Claude just killed my startup." My first reaction: that's brutal. My second reaction: I've seen this before and it wasn't Claude that killed the product. That is a recent story. I watched a team spend months building a Slack bot that

Sam Altman says don't learn to code. I think that's the most dangerous career advice in tech right now. He says you need "high agency, soft skills, idea generation" instead. Sure, he is right. But here's what he's missing. Every time a new abstraction layer appears, the people

Imagine you're a CTO with a government contract. Your product runs on Claude. Monday morning, you learn Claude is banned from all government systems. Now what? If you built your AI layer as a plug-and-play abstraction (model behind an interface, swappable with a config change)

Block laid off 40% of the employees last week. That's not an AI story. There are people out there blaming AI for every layoff right now. But what actually happened is simpler and older than any LLM. There was a period money was nearly free. Interest rates were on the floor.

A missile hit a cloud data center a couple of days ago. Your disaster recovery plan just became a board-level conversation. Yesterday, a major AWS data center went offline. The official status page said "localized power issues." Here's what I know from running production

The creator of Claude Code just said the "software engineer" title is going to disappear. I disagree but also agree ;) I've been building software for over 20+ years. I've watched titles come and go. "Webmaster" became "frontend developer." "DBA" became "data engineer."

18,000 people liked a tweet saying Tony Stark was a vibe-coder. He wasn't. He understood arc reactor physics before he told JARVIS what to build. The engineers who are 'vibe-coding' successfully aren't vibing. They're domain experts who happen to prompt instead of type. The ones

Will vibe-coding kill SaaS? No it won't. And here's why. Marc Benioff just revealed that OpenAI and Anthropic are among Salesforce's top customers this quarter. Think about that for a sec. The companies building the AI that's supposed to "replace SaaS" are themselves paying

One year ago this week, Dario Amodei said AI would write 90% of code within 12 months. Everyone laughed. He was right. And he was wrong. At startups and AI labs? It's 90%+. Maybe higher. Cursor, Claude Code, GPT Codex, they're writing entire features end to end. Engineers are

Amazon now requires senior engineer approval before AI-written code goes to production. My first question: how did they ever ship without it? A 6-hour outage on Amazon.com was traced back to an AI-assisted deployment. Not a complex distributed systems failure. Not a

Claude went down this week. I saw developers posted: "I guess I'll write code like a caveman." That sentence should terrify every CTO. Not because AI tools go down. Everything goes down. We would never let production depend on a single database. We build replicas, failover,

Google turned Google Maps into an AI agent. 2 billion users. Every local business on earth. And now an AI model sits between them. I think there is something we need to understand here: this isn't a feature update, this is a platform shift. Every local business just got a new

"Are you a data engineer, a data scientist or a backend developer?" Someone asked me this recently. Data engineer? Data scientist? ML engineer? DevOps? Backend developer? I said: "I'm an engineer." They looked confused. Here's what I've noticed, ONCE AGAIN. Somewhere along

$150 million. For a cancel button. Adobe just agreed to pay $150 million because they made it too hard to cancel a subscription. I don't use Adobe. But I've lost count of how many times I've read a complaint about Adobe's hidden fees, buried cancellation flows, etc ... I

Most people learn ML by collecting random tutorials. That's like learning to cook by watching 50 different YouTube channels and never making a full meal. If you're building end-to-end ML systems, study them systematically. Not just the model. The pipeline. The deployment. The

Terence Tao, one of the greatest living mathematician, says the math behind LLMs is simple. He's right, but here's what people should also ask: If the math is this simple, why can only 5 companies on Earth build frontier models? It's not the equations. It was never the

Claude Code wiped a production database with a single Terraform command. 2.5 years of submissions, homework, projects, leaderboards and all gone. My first question: who gave an AI agent Terraform access to production? Not "why did the model hallucinate."  We don't give junior