From Prompt Engineering to Outcome Engineering

We’re moving from a bizarre phase of companies using “tokenmaxxing” as an indicator of employee AI adoption, a wasteful metric easily gamed, to “outcomemaxxing” as an actually useful measurement approach.

Twenty-dollar subscriptions are fine for many use cases, but as increasingly capable (and expensive) thinking models like Anthropic’s new Fable are integrated into business practices, the costs for highest-skilled tasks need to be clearly defined. Measuring completed work will be crucial for companies to budget token use effectively and determine whether their AI investments are delivering solid ROI.

EXAMPLE: Yesterday I burned through a quarter of my weekly Claude Max token allowance in only FOUR prompts using Fable. Yes, the prompts were incredibly detailed with one taking me about an hour to craft. In total the AI churned for about 90 minutes.

OUTPUT: While expensive at first glance, the output was equivalent to ~3 days work of an expert data analyst and probably about ~4 days’ work of a senior developer.

TAKE AWAY: The final deliverables were of excellent quality. When properly orchestrated even the most expensive AI models can deliver an incredible ROI.