Summary
Craig Hewitt, founder of podcast platform Castos, delivers a grounded take on AI adoption drawn from consulting with more than a dozen companies ranging from 20 to nearly 5,000 employees. Rather than focusing on job cuts, he argues that organizations are pursuing AI in three sequential phases: first, increasing throughput (more marketing, more product, more customers served); second, improving quality using richer data and feedback loops; and only then reducing costs — including headcount, which he sees shrinking by roughly 10% annually at some larger clients, but more slowly than the headlines suggest.
One of the more specific observations in the video concerns enterprise AI pricing: Claude’s enterprise plans charge per token on top of seat licensing, while OpenAI and Codex bundle token allowances into the seat cost. For large organizations running heavy agentic workflows, this difference can be substantial, and Hewitt argues that cost-conscious buyers should examine it carefully rather than defaulting to Claude based on reputation alone. His own six-person team spends over $1,000 per month on AI tooling, a figure he notes scales dramatically at enterprise headcounts.
Hewitt’s broader argument is that the predicted AI job apocalypse isn’t materializing because capacity expansion in phases one and two creates enough business value that headcount reduction becomes neither urgent nor necessary in the near term. The video is informal — recorded on AirPods while traveling — but the content is substantive and grounded in direct operator experience rather than punditry.
📺 Source: Craig Hewitt · Published July 30, 2026
🏷️ Format: Opinion Editorial
