
AI as a variable in the system

Signals & Levers: Systems Thinking to Navigate Software Delivery Illusions - Elisabeth & Joel
Topic 9 of 12
Published Aug 3, 2026 · Clip 50:32–1:01:12 (10:40)
Elisabeth says AI is real and transformational—she partners with Claude daily—but the right question isn't whether it's useful, it's 'what is the best way for us to apply this technology in our context,' and answers change every couple of months, making signals and levers more important than ever. Joel explains reinforcing and vicious loops: introducing new technology (platform teams, cloud, AI) can flip a virtuous loop into a vicious one, as with platform teams abstracting Kubernetes until teams can't deploy. If leadership doesn't understand the system, AI can exacerbate existing problems—thousands of AI-generated lines into an already hard-to-reason codebase will hurt, while simple architectures with clear boundaries and testability let AI amplify. They pose the open question: what happens when you ship faster than you learn? Elisabeth shares that the best approach is optimizing for learning—describing a company that ran a week-long 'vibeathon' purely to learn AI together—admitting quarterly hack days are no longer enough and predicting harnesses orchestrating many agents will define software development; Joel adds learning should be paired/group-based and AI adoption treated as a hypothesis with shorter feedback loops and countermeasures.