Cursor's Figma MCP means I stopped eyeballing screenshots for spacing and color — I get the real node data. And nothing ships until I have watched it live on an iPhone plugged into my Mac, because Safari, not desktop Chrome, is the browser most of my clients' actual customers use.
Not a hype post about "10x with AI." A concrete walkthrough of where AI actually sits in my daily work as a full-stack developer — code, content pipelines, translations, tests, deploys — and where I still do the work myself.
The most expensive Python I review is for order in orders after a SELECT *. Last row per group, month-over-month, filtered counts, and upserts are queries. How I write them in Postgres instead of downloading the table.
The SQL can be right and the API still dies: unbounded .all(), f-string queries, ORM N+1, a new connection per request, pandas as a warehouse on the web box. How I keep Python thin around Postgres in 2026.