Claude Code

In December 2025 a colleague introduced me to Anthropic’s Claude Code. It worked the way I think: in a terminal, against the real repo, with the shell and the tools I already used.

Within two weeks I had downgraded my Cursor subscription and bought Claude’s top-tier plan. It became, and remains, my preferred pairing partner.

The difference was not the model on its own. It was that the agent could run things: read the cluster, run the tests, check the logs, and come back with evidence. For someone who has spent a career insisting on evidence before action, that was the unlock.

What changed: pairing went from autocomplete to delegation with verification.