The "manage it like a coworker" framing works until you ask: who is the coworker, exactly? I have recently been thinking about the issue of Agent identity. Your human colleague has a reputation, a contract, an email address. When an agent acts through your account, the outside world sees only you. Audit trails help, but they're a workaround for a problem that hasn't been solved at the infrastructure level yet — agent identity. Until that exists, you're not really managing a coworker. You're co-signing everything it does.
The 20% reframe is the most useful thing in a piece full of useful things — taking a number that sounds underwhelming and doing the actual arithmetic of 100 hours a year is the kind of grounding most AI writing deliberately avoids. Just curious to know, how do you personally draw the line between offloading and atrophying?
The "manage it like a coworker" framing works until you ask: who is the coworker, exactly? I have recently been thinking about the issue of Agent identity. Your human colleague has a reputation, a contract, an email address. When an agent acts through your account, the outside world sees only you. Audit trails help, but they're a workaround for a problem that hasn't been solved at the infrastructure level yet — agent identity. Until that exists, you're not really managing a coworker. You're co-signing everything it does.
I believe that AI coworker is not there to replace your “human coworkers”, but rather, it is there to augment your skills and productivity.
The 20% reframe is the most useful thing in a piece full of useful things — taking a number that sounds underwhelming and doing the actual arithmetic of 100 hours a year is the kind of grounding most AI writing deliberately avoids. Just curious to know, how do you personally draw the line between offloading and atrophying?
🔥🔥🔥