Situational Leadership matters just as much with AI projects as with human teams...great managers don't treat every member of the team the same way. They don't interact with the same frequency, or provide the same level of guidance or oversight.
In practice with AI, this might look like a process superset used across tools, with different levels of instruction, validation, task granularity, frequency of interaction/intervention, and specificity of task used depending on the tool, the environment, and the task.
In the past 40+ yrs, I've seen sw mgmt practices evolve to allow devs to apply creativity to hard sw problems r/t reinventing mgmt processes each time. @AlabamaMike points out that nuance drives which process ("Outer Loop Harness") is best for Agentic Coding/Context Eng, too.
Looking to deploy agentic coding in your dev shop? Check out my latest article: Learning About Outer Loop Harnesses for Agentic Coding: What CTOs Need to Know https://t.co/I6MXWwbln4 via @LinkedIn
Joe Magerramov's recent post (https://t.co/DSnZxBhX9a) raised an idea that I haven't thought enough about in my exploration of spec-driven devt & agentic coding. He didn't give it a name, but I'll call it "synthetic testing environments", an extension of "synthetic testing data".
Better lucky than good? Maybe "past performance doesn't guarantee future success", but I'll always bet on skill PLUS a bias for action plus grit enough to stay in the game until lucky opportunities arise! | BBC Capital #success#csuiteskills#luckvsskill https://t.co/8ZDGNtb6ri