Notes from the work: what we tried, what broke, and what we'd do again.
Spirit Airlines went bankrupt. Google bought its 100 million emails and 500 million Teams chats to train AI. It's a bargain, a privacy fight, and a clear sign that the labs are running out of data.
Project Glasswing found a 27-year-old vulnerability in OpenBSD and a 16-year-old flaw in FFmpeg that five million automated tests missed. This isn't a research preview. It's a signal about what AI is capable of now.
Anthropic accidentally shipped Claude Code's source code in a release package. A hacker targeted the axios NPM library to push malware to millions of projects. Two incidents, one root cause — and a clear signal for teams building with AI.
AI lets you build at unprecedented speed, but speed without guardrails means more bugs in production. The fix isn't slowing down. It's letting AI write your tests too.
We found a gap in the market — client portal software was either overpriced SaaS or a patchwork of tools. So we built the one we actually wanted to use.
AI agents lived in demos for years. In 2026, they're finally shipping in production. Here's what changed, what's actually working, and where most teams are going to get it wrong.
A few weeks ago nobody had heard of it. Now OpenClaw is one of the fastest-growing open-source projects in history. Here's what it is and where it's headed.
We take on a handful of client projects a year, usually with small teams that need an AI feature, an automation, or a working prototype fast.