CC-201c · Module 1

When to Compose vs Keep Simple

3 min read

There is a gravitational pull toward complexity once you learn composability. You can chain skills, spawn sub-agents, pipe results through MCPs, and orchestrate multi-step workflows. The temptation is to build elaborate automation for everything. Resist it. Over-engineering AI workflows is just as real and just as wasteful as over-engineering software. A single well-crafted prompt that takes thirty seconds to type will outperform a Rube Goldberg machine of three skills and two sub-agents for most tasks.

The decision heuristic is frequency times complexity. If you do something daily and it involves five or more steps, compose it into a skill. If you do something weekly and it involves three steps, a prompt template in your notes is probably sufficient. If you do something once, just type the prompt. The threshold for automation should be honest: would creating this skill save me more time than it costs to build and maintain? Most developers overestimate the frequency of their workflows and underestimate the maintenance cost of their automation.

Do This

  • Automate daily workflows with five or more steps (deploy, CI, daily audits)
  • Use a single prompt for one-time or low-frequency tasks
  • Keep skills focused — one skill, one job
  • Delete skills you have not used in a month

Avoid This

  • Build a three-skill chain for something you do once a month
  • Create a sub-agent pipeline when a single prompt would work
  • Add MCP servers "just in case" — every MCP costs context tokens
  • Keep unused skills around "because I might need them" — dead skills clutter context