Right now I feel like I have superpowers at work. It confuses me very much when I hear about engineers and scientists who are hesitant to fully adopt/integrate frontier models into their daily worksteams.
I realize adopting AI workflows requires a fundamental shift in how we think about frontier model capabilities. We as the user of AI have to take responsiblity for when an AI fails to complete the intended task. When someone says "AI doesn't work," it's almost always because the prompt lacked the complete context needed for the task. When a model "doesn't do what I expected," it's almost always because the model filled in gaps left by an underspecified task.
The other barrier is verification. When an experiment produces results or a feature is fully built, how do you check them? My approach is to generate a visual representation at two levels — a 30,000-foot view and a ground-level view. When both are enabled, the AI produces a consistent view anchored between the dashboard numbers and the individual record results. If you understand what drives those rollups, you can verify all the dependent parts automatically.
These two tenets — complete context and structured verification — are what allow you to move at the speed of AI.
The title plays on the tagline from a16z crypto's "AI just gave you superpowers — now what?".