Bobby Smart

Updated July 2026

The stack I actually run

What I pay for and use most weeks, and what it's genuinely for. I buy these myself, so the verdicts are mine. No vendor relationships, and no price tags here because they change faster than pages get updated.

  • Claude Code

    the big plan, paid monthly

    My terminal agent for the heavy lifting: planning features, reasoning about a whole codebase, and big multi-file refactors. Plan mode before code is the habit that pays off. I'm on the top-tier plan and it only makes sense because I lean on it every day.

  • Cursor

    paid plan

    The IDE I live in. Composer for multi-file edits, and its tab completion is the best I've used. Agent mode is powerful but needs watching; leave it unsupervised and it'll cheerfully write CSS modules into a Tailwind project. I came from VS Code and haven't looked back.

  • Codex

    paid plan

    OpenAI's coding agent, and the second opinion in my workflow. When two agents plan the same feature and trade their homework, the disagreements are where the real design decisions hide. It also carries its share of the review work across my repos. Different instincts from Claude, and that's exactly the point.

How I work with them

The one trick I run constantly: dual planning. I hand the same task to Cursor's Composer and to Claude Code, let each write a plan, then trade so each reviews the other's. The merged plan beats either draft, and I wrote up why in the notes.

Beyond that: start with the smallest thing that works and add complexity only when it earns its place; avoid third-party packages where a small util will do (and vet anything that stays); and keep feature branches even on solo projects, so an agent's bad afternoon never lands straight on main.