Artificial intelligence

I shipped more than I ever had. The output was real, the speed was real, and something underneath was rotting that I did not notice for months, because the dashboards only measure the part that got better.

I stopped treating Claude like Google. That one shift changed everything.

Five real Python tasks. One model upgrade that actually changes the comparison. Here is what happened.

Most small engineering teams don’t fail at Kubernetes because it doesn’t work.

LLM inference failures don’t look like normal backend failures. They look like nothing until they look like everything.