Blog
What I learn building with AI as a business-side Ops.
What changed when I stopped choosing a model per repo
Earlier in 2026 I kept Codex and Claude apart, each with its own directory and its own job. They now work on the same files, and I cannot tell their output apart. What changed is the repo they work in, not my judgement of either tool.
One definition of a metric, even for the AI
When you let an AI query your numbers, the real risk is whether it agrees with everyone else on what the numbers mean. A semantic layer is the answer, and I tested it on my own training data.
AI-first does not mean workflow-free
Not everything should live in a Claude conversation. n8n brings structure, triggers, and auditability to processes that need to run reliably. Its agent nodes make it more powerful, not less relevant, in an AI-first world.
Not everyone vibe codes equally: 7 challenges for business-side teams
The barrier to building with AI has never been lower. The barrier to doing it well has not moved. Here are seven challenges that business-side teams may underestimate when deploying AI.
I tried to keep Codex and Claude apart on the same project
Codex read repos and spotted problems, Claude Code fixed them, and each worked from its own directory. This is how that ran through the first half of 2026, and where it held up.
First steps in Claude Code as a non-developer
The installation takes twenty minutes. The mental shift takes a few sessions. This is exactly what to install and how to structure your first project if you have no development background.
Why I use Claude Code over Cowork
Claude Cowork was built for people like me. And yet Claude Code is the tool I open every morning. Here is why.