Notes from
the field.

Everyone's blaming North Korea for the $6.3bn Binance heist, and they're probably right. But it's worth asking what a rogue AI would do for money.

A CTO friend can no longer pick out the 10x engineers on his team, because agentic coding has turned every capable engineer into one.

Stripe already owns stablecoin issuance, wallets, and a chain. Buying PayPal would put 438 million funded accounts one app update from Web3.

Fixed Temporal workflow IDs, an UPDATE that doubles as a lock, and fail-closed reconciliation: layering double-charge protection into a payments system.

With agentic teams, two hackers can spin up five MVPs while a normal team ships one feature. The case for keeping a skunkworks in your back pocket.

Coding agents should run unsupervised, but inside a dev environment that's safe to be wrong in. The agent boundary is the wrong line to defend.

Most companies run a migration campaign to clear users off a deprecated API. Anthropic just tripled the price of theirs and let market forces do the work.

GitHub had 3,800 internal repos exfiltrated via a poisoned VS Code extension. If your private repos have ever held a secret, rotate it today.

Brent Roose says PHP's biggest problem is its marketing. The same is true of almost every business that's quietly good at what it does.

OpenAI is reportedly burning $1.69 for every dollar earned, and Anthropic is already raising prices. If your AI business is mostly API markup, the price floor is coming for you.

Google shipped multi-token prediction for Gemma 4. The headline 3x speedup is for big GPUs, but the edge-model drafters quietly target laptops like mine.

Engineers treat coding agents as smart junior devs. Vibe coders treat them as their CTO. The gap matters more than the AI hype suggests.

What this section is for, what to expect, and the loose rules I'm writing it under.