Stop Looking for the Perfect Prompt: The Design-First Workflow for Coding Agents
summary
Argues the "perfect prompt" is a myth and prescribes a design-first, stage‑one workflow: write a Markdown spec as external memory, use Cursor/Claude Code planning mode and downloadable skills to produce a reviewable development plan, then run subtasks and review diffs. Demonstrates replacing a scattered --fusa flag with a generic policy mechanism in an automotive image builder, shows using sub-agents, focused compaction (slash compact), and promotes decisions into docs rather than letting context drift.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 3:10Spec-driven infrastructure
- 8:06Stage one explained
- 14:51Context window and drift
- 19:22Executing and reviewing
- 23:34Q&A: TDD and spec clarity
- 26:05Q&A: Incremental vs batch review
- 28:13Q&A: Adopting in projects
- 29:35Q&A: Iterating specs
- 31:39Q&A: Archiving specs
related talks
48mStop Fighting Your AI Engineering Prompts That Actually Work
44mSupercharge Your Productivity Mastering Prompt Engineering
29mSupercharge Your Productivity: Mastering Prompt Engineering
29mWriting Less, Owning More: The Shift in Technical Writing
22mNavigating the Evolution of Prompt Engineering
45mAI-powered Code Development: The role of prompt engineering
46mPractical End-to-End AI Development using Prompty and AI Studio
01h 35mBuilding with Anthropic Claude: Prompt Workshop