Gary Tan’s journey—from investor to hands-on engineer—illustrates a broader shift: AI tools like Claude Code and OpenClaw aren’t just accelerants but true collaborators. He rebuilt Posthaven in five days using agentic workflows, while GarysList.org serves as both a civic journalism platform and a live lab for 'tokenmaxxing'—feeding rich context to AI for sharper research and coding decisions. The discussion highlights that tool choice reflects cognitive style (e.g., Claude Code for rapid ideation vs. Codex for architectural rigor), and stresses that human judgment remains central for QA, intent-setting, and verification. Developers are now composing systems of AI agents—like GBrain’s RAG-powered newsroom—to achieve 400x productivity gains over a decade ago. Yet efficiency isn’t about speed alone: it’s about personal agency, prompt authorship, and treating token budgeting as a strategic investment. The future isn’t AI replacing coders—it’s individuals wielding composable, open, user-controlled agents to build meaningful things, fast.