What's this about?
Ethan Mollick says current Astra and Fable models can already carry out work that would take people weeks if they are guided and their output is harnessed correctly.
What exactly is a model harness?
It is the surrounding tools, instructions, memory and checks that let a model act reliably—like the workspace, checklist and supervision around a capable new employee.
Why is it worth attention?
It shifts the question from waiting for a smarter model to reorganizing work around today's models. It remains a prominent practitioner's observation, not a measured economy-wide productivity result.