Teach Your AI to Think Like a Senior Engineer

Klaassen presents eight concrete planning strategies organized by complexity level, each implemented as a specialized research agent. The key concept: running parallel research agents before writing any code prevents building the wrong solution entirely, while the human contributes judgment and taste.

Strategy highlights include: reproduce-and-document (diagnose before fixing), ground-in-best-practices (web research agent), and ground-in-codebase (prevent reinventing existing solutions). Each strategy includes a compound step where findings are saved to project docs, building an ever-growing knowledge base that agents consult before searching externally.

The article includes GitHub links to the exact agent definitions and commands used, making the system reproducible. The email bankruptcy example recurs: a research agent’s 20-minute investigation of Gmail rate limits prevented days of wasted implementation.

RDCO mapping: The parallel-research-before-action pattern maps directly to the RDCO research brief workflow. The strategy of saving agent findings to persistent docs mirrors vault contribution. The eight strategies could be adapted as an operational checklist for any complex task where upfront research prevents downstream rework.