In this episode of Shipping Excellence, Andrew Duncan sits down with Christopher Garvis (former AWS engineering leader) to unpack what changes when AI compresses software delivery timelines.

They explore why “skills” (constrained context plus tools) are emerging as a more reliable pattern than general-purpose agents, when MCP-style integrations become heavy or fragile, and how experienced builders are using parallel AI workflows to increase throughput without sacrificing rigor.

The conversation then moves from tooling to fundamentals: if AI makes execution cheaper and faster, what becomes scarce? Judgment, taste, ownership, and integrity. They discuss why AI amplifies whatever discipline you already have, why “vibe coding” hits a wall at security and reliability, and why follow-through may be the true moat as software becomes easier to replicate.

Topics covered - Parallel AI workflows in the terminal (multiple concurrent instances) - Skills vs agents: constrained context, defined tools, clearer outcomes - MCP servers in practice: tradeoffs, token cost, latency, brittleness - When to use CLI tools vs direct commands inside a skill - Emergent behavior: models adapting how they use tools - AGI and embodiment: what text-only systems miss about being human - Why AI increases the value of judgment, taste, and decision-making - How AI amplifies process quality (and exposes weak discipline) - Vibe coding: when it’s fine, and when you must harden the system - The new moat: ownership, integrity, and doing right by customers

Guest Christopher Garvis (VP of Engineering and Delivery at Vertice Labs)

Links Guest: https://www.linkedin.com/in/christophergarvis/ Duncan: https://www.linkedin.com/in/andrew-b-duncan/ Vertice Labs: https://verticelabs.io