Create an account to save sources and durable progress.
Lenny's Newsletter · Claire Vo
podcastplaybookSource excerpt
About Lenny's NewsletterWatch now (24 mins) | 🎙️Meta launched Muse Wonder with claims of frontier agentic performance, so I ran it headlessly through Claude Code against nearly 20 tasks to find out if any of it is true
This is a limited feed-provided excerpt, not the full original work.
Apollo layer
Evaluate AI models against the distinct jobs in your workflow rather than treating overall capability as interchangeable. A model that performs well on agentic tasks may still be a poor choice for design work.
Adopting a model based on broad performance claims can lower output quality in specialized tasks. A small, task-specific evaluation helps you choose the right model—or combination of models—before embedding it in your product or operations.
List 10–20 representative tasks for each critical workflow, define what a good result looks like, and run candidate models under the same conditions. Which model should own each task based on observed quality rather than headline claims?
Private notes
© Apollo Advisors° ABN 34 346 108 139