From Consulting AI to AI That's Already Finished the Work

This episode picks up a thread from the day before — the observation that major tech companies are now spending as much on AI tokens as on engineer salaries — and uses it as a starting point to look at why so much corporate AI adoption produces no measurable result.

The central finding, drawn from a report on AI in corporate finance departments, is striking: 95% of AI rollouts have had no impact on the bottom line. The episode works through why that might be, and lands on a simple structural problem — when AI requires a human to type a prompt every single time, the workflow hasn't changed, only the speed of individual steps within it.

The contrast drawn is between that prompt-each-time model and something called background agents: AI systems that trigger automatically from events, like an invoice arriving, and complete the routine work before anyone sits down at their desk. An example from the finance world makes this concrete — by the time a person arrives in the morning, only the items that genuinely need a human decision are waiting.

There's also a candid look at how this applies to running a small optical shop, where tasks like cross-referencing appraisal prices, converting lens prescription data, and syncing inventory across platforms each take only a few minutes but together occupy a significant part of the day. The line being drawn isn't about handing everything to AI — it's about identifying exactly where the boundary should sit between what AI handles automatically and what stays with the person who knows the work.

A quiet reflection on the difference between AI as a tool you consult and AI as something that quietly finishes the work before you arrive — and what it might mean, for a small shop, to get that distinction right.