When in Doubt, Experiment—Be Stubborn on Vision, Flexible on Details

Throughout their careers, Bill Carr and Colin Bryar witnessed firsthand how unwavering commitment to a clear vision made ambitious projects possible, even as the details demanded constant adjustment. Early in their Amazon days, they saw attempts to implement complex team fitness functions and tightly-scored metrics—each a creative solution but sometimes more distracting than clarifying. Team meetings often became debates over metric weightings instead of execution.

Instead of locking into these experiments, the most successful leaders rapidly tested on small scales, ready to revise or abandon details that weren’t serving overall goals. The two-pizza team idea wasn’t perfect and didn’t fit every context, so it was allowed to morph into single-threaded leadership, preserving the core principle (autonomous, focused teams) while changing structure and expectations.

Big wins came where teams showed intellectual humility, updating their approach with data, reflection, and sometimes external feedback. Teams would pause, run retrospectives, and pick what to keep or cut. Being flexible in the 'how,' while stubborn in the 'why,' transformed stalling projects into thriving engines.

Behavioral science tells us experiments teach faster and safer than betting everything on unproven systems. By protecting the vision—but letting the tactics shift—leaders progress resiliently and learn faster.

Put your North Star or the most important goal somewhere visible and declare it with your team—make sure everyone knows what success looks like in the long run. Pick a process you’ve wanted to improve (hiring, weekly meetings, project launches) and try a new approach with a small group or on a trial basis. At the end, gather everyone involved and ask honestly what surprised, frustrated, or delighted them; then, revise your process based on their feedback. Keep what fits your vision, but don’t be afraid to scrap what doesn’t. Approach improvement as a series of safe-to-fail experiments, rather than all or nothing.