Quote · This Week in Tech (Audio)
TWiT 1093: California Sober - Kimi K3, Qwen3.8, & China's Open-Weight AI Gambit
Where this was said
Agentic Coding Workflows: YOLO Mode, Loops, and Multi-Agent Review Pipelines
At 29:44 · chapter starts 19:15
This chapter is the most technically dense of the episode, as Harper Reed and Leo Laporte compare notes on their actual agentic coding setups. Harper describes using '/loop' and '/goal' commands to keep agents iterating until tests pass, then passing output to a second model for adversarial review — what he calls 'giving it more loops.' Leo describes a more elaborate multi-agent pipeline: Claude Fable plans the work, Opus 4.8 does the coding, Sol 5.6 reviews it, Grok 4.5 does a final review, with agents communicating via Obsidian text files acting as a shared mailbox. Harper notes his company now writes significant amounts of Rust, Swift, and Go — languages they never would have used manually — because the agent doesn't care. Alex Wilhelm makes the memorable observation that children born today will likely never hand-write code in languages like C++, and that manual coding will become a lost art.
Waymo is statistically 10–15x safer than human drivers, yet public skepticism persists — a pattern the panel sees repeated in debates about AI-generated code quality.
Multi-agent review pipelines, YOLO mode, sub-agents that email each other test results — Harper Reed's 6-person company has built an AI workflow that generates more output than his entire prior career. The secret isn't one big model; it's a harness that routes work to cheap models and uses premium ones only for planning and review.
Nobody wants to write Rust by hand. But when your AI agent writes it for you, there's no reason not to use the safest language available. Harper Reed's company now ships significant Swift, Rust, and Go code — languages they'd never have used before — because the agent doesn't care what language it uses.
