It’s Wednesday, September 23, 2026. You’ve got the good blanket, the remote is somewhere in the couch cushions, and you’ve just pressed play on Brothers on Apple TV. Matthew McConaughey appears. Woody Harrelson appears. And then a strange little thought lands: neither of them is exactly playing a character. They’re playing fictionalized versions of themselves, whose lifelong friendship gets thrown into chaos.
I write about AI agents for people who don’t build them, and I sat with that idea longer than the show probably intended. Because “a fictionalized version of you, acting on your behalf, occasionally causing chaos” is also a fairly accurate description of what an AI agent is.
What an AI agent actually is, minus the jargon
A chatbot answers you. An agent goes and does things. That’s the whole difference, and it’s a big one.
If you ask a chatbot to plan a trip, you get a list. If you ask an agent, it may check your calendar, compare flights, fill out a form, and send an email signed with your name. It’s operating in the world with your identity attached. Nobody watching that email land in their inbox knows a model wrote it. They just see you.
So the agent becomes a fictionalized version of you. It sounds like you, more or less. It makes choices you’d probably make, most of the time. But it is a performance of you, assembled from your instructions, your past messages, and some guesswork in the gaps.
The gaps are where the plot happens
McConaughey and Harrelson have decades of actual friendship to draw on, and the premise still turns on things going sideways. That’s a screenwriting choice, but it maps neatly onto the real risk with agents: the interesting failures aren’t the obvious ones.
An agent rarely goes wrong by doing something wildly out of character. It goes wrong in the small stuff, the places where it had to improvise:
- It replies to a message with a tone you’d never use with that particular person
- It books the cheaper flight because you said “save money,” not knowing you meant “within reason”
- It assumes a yes where you would have paused and asked a question
- It follows your instruction perfectly, and the instruction was the problem
None of those are dramatic betrayals. They’re the small drift that happens when a stand-in fills in your blanks. Watch enough of it and you start to understand why “give the agent more autonomy” is a bigger decision than it sounds.
Why the self-portrayal framing is useful
Most explanations of AI agents reach for the wrong metaphor. People say “digital assistant,” which makes you picture someone taking dictation. Or they say “employee,” which implies judgment and accountability that software doesn’t have.
“Actor playing a fictionalized version of you” is closer, and more honest about the tradeoff. An actor playing you needs direction. They need to know what you’d never say, which relationships are delicate, where the lines are. Without that, they’ll deliver a plausible performance that anyone who knows you would find slightly off.
That’s exactly what setting up an agent well requires. Not clever prompts, but context. The boring, specific, unglamorous details of how you actually operate.
What to do with this, practically
If you’re experimenting with agents at work or at home, borrow the director’s mindset rather than the boss’s.
Give notes on what went wrong, specifically. “Too formal with Dana, she’s a friend” is more useful than “be better.” Decide in advance which scenes the agent plays alone and which ones you’re in the room for. Anything involving money, hiring, public statements, or someone’s feelings is worth keeping on your side of the line for now.
And check the output before it ships, at least early on. Not because the technology is untrustworthy, but because you’re the only one who knows whether the performance rings true.
Enjoy the show, then think about the inbox
Apple TV held the global premiere at the Hammer Museum, with Holland Taylor among the attendees, before the series debuted worldwide. It’s a comedy about friendship, and I’d rather it stay that way than be recruited as a tech parable.
Still, the premise is a genuinely good teaching tool, and those are hard to find in this space. Two people, known quantities to each other, playing slightly-not-themselves, and the friction that comes from the gap between the real thing and the version on screen.
That gap is the whole story with AI agents too. Mind it, direct it, and you get something useful. Ignore it, and you get comedy, except it’s happening in your actual life.
đź•’ Published: