The Context Window
Jeff Escalante (Clerk) on Codex, Agent Costs, T3, and Why Humans Still Need to Review the Code
Brandon Mathis talks with Jeff Escalante of Clerk about what it actually takes to go all in on AI-assisted software development. They compare Codex and other coding tools, examine why subscription pricing changes the economics of agentic work, and make the case for keeping humans firmly in the loop—especially for security-critical products. The conversation also explores AI code review, productivity signals, and why today’s project-management habits are struggling to keep up with a multi-agent workflow.Explore Clerk at clerk.dev and join the conversation on how AI is reshaping practical engineering work.5 Key Takeaways- Codex is a practical daily driver for AI-assisted development, but the real value comes from active human direction—not simply handing work to an agent.- Subscription plans can make intensive AI coding far more economical than API-priced cloud agents; high token usage can represent thousands of dollars of theoretical monthly spend.- Unmonitored agent swarms are not dependable enough for security- and reliability-sensitive systems. Authentication platforms such as Clerk cannot afford autonomous mistakes.- AI can meaningfully increase engineering output, including drafting and reviewing code, yet its work still requires close reading, frequent course correction, and human accountability.- Traditional productivity metrics and project-management systems are increasingly mismatched with a workflow where engineers supervise multiple concurrent AI-driven projects.Brandon Mathis on Linkedin: https://www.linkedin.com/in/mathisbrandon/ Jeff Escalante on Linkedin: https://www.linkedin.com/in/tracyslee/ This Dot Labs Twitter: https://x.com/ThisDotLabs This Dot Media Twitter: https://x.com/ThisDotMedia This Dot Labs Instagram: https://www.instagram.com/thisdotlabs/ This Dot Labs Facebook: https://www.facebook.com/thisdot/ Sponsored by This Dot Labs: https://www.thisdot.co/
