What's new in AI?

John Maeda shares a practical take on what’s genuinely new in AI: set a goal, define how you evaluate proximity to that goal, and let the system improve via an iterative loop.

Overview

• Define a goal: Start by being explicit about what you want the system to achieve.
• Define evaluation: Specify how you will measure “proximity” to the goal (the success metric).
• Iterate in a loop: Let the system hill climb toward better outcomes through repeated evaluation and improvement.
• Why this works now: Capabilities that seemed possible a decade ago now work in practice because tokens are cheaper and more powerful.
• Core idea: Build compounding improvement systems—systems that get better through repeated cycles of measurement and adjustment.