Thariq Shihipar, an engineer on Anthropic's Claude Code team, cut Claude Code's system prompt by 80%. That single fact reframes everything people assume about prompt engineering for agentic systems, and it anchors a 40-minute walkthrough covering his full build workflow, from planning to parallel execution.
The substance is in the method. Thariq uses a /goal command to keep Claude running in extended loops without losing context, plans by explicitly removing unknowns before writing a single line of code, and runs parallel Claude agents coordinated through Slack. He also separates creator and verifier sub-agents, a architecture decision that directly addresses the reliability problem most people hit when they try to automate beyond simple tasks. Anthropic uses HTML artifacts internally as planning documents, not just output formats.
The 80% system prompt cut is worth the read on its own, but the deeper argument is about what makes someone more technical in an era where AI writes the code. Thariq has a specific answer. The live demo of generating a polished launch video from a single prompt gives the workflow a concrete anchor. Full episode is 40 minutes and runs through every layer of his stack.
[WATCH ON YOUTUBE →]