Building with AI coding agents

How to plan, track and finish software with Claude Code, Cursor, Codex and the rest: written from running them every day.

• Claude Code

  How to give Claude Code memory across sessions

  Claude Code starts every session fresh. What it remembers (CLAUDE.md, resumed chats), what it forgets, and how to keep your plan where every session reads it.

  7 min read · Updated Oct 21, 2026
• Rules files

  CLAUDE.md vs AGENTS.md: rules files for coding agents

  What CLAUDE.md, AGENTS.md and the other agent rules files do, which tool reads which, how to keep one source of truth, and what belongs in them.

  6 min read · Updated Oct 21, 2026
• MCP

  Give your coding agent a task board over MCP

  How an MCP project board works with Claude Code, Cursor and Codex: what the agent reads and writes, how it differs from an issue tracker, and how to set one up.

  7 min read · Updated Oct 21, 2026
• Multi-agent

  Running several coding agents on one repo without collisions

  Run Claude Code, Cursor and Codex in parallel on one codebase: git worktrees, splitting work by feature, claims, dependencies and reviewing the results.

  6 min read · Updated Oct 21, 2026
• Planning

  Writing acceptance criteria for AI coding agents

  Write stories coding agents can finish: checkable acceptance criteria, the right story size, examples for UI, API and bug fixes, and what backfires.

  6 min read · Updated Oct 21, 2026
• Vibe coding

  A vibe coding workflow that actually ships

  Vibe coding gets a demo working in an evening. This workflow gets it shipped: one-sentence features, small stories, build by ID, review, and resume any time.

  6 min read · Updated Oct 21, 2026