After twelve years of shipping software, the fundamentals I care about are simple: owning projects from start to finish, shipping in small, reversible steps, building reusable foundations instead of one-offs, and instrumenting problems to find the actual root cause.
Having those habits in place is what lets me get real leverage from AI. I've built my day-to-day workflow around coding agents, with my codebase, issue tracker, and pull requests connected via MCP and several agents running in parallel. For larger work, I iterate on a detailed spec with the model before any code gets written, and I set up portable repo rules and skills (AGENTS.md, SKILL.md) so agents follow our conventions.
Working this way has multiplied my output and made it much faster to jump into unfamiliar parts of the stack. Agents handle most of the implementation and help map new territory, while I focus on the design decisions, catch bad assumptions, and review every line that ships.