What is wrong with us?!
TL;DR
AI agent culture has transformed Bay Area tech energy from raw optimism about opportunity into a perpetual anxiety about falling behind -- the pressure never decreases, it only multiplies with each new model release and capability jump.
The ease of spinning up AI-coded projects is a Faustian bargain: you can build more than ever before, but the work expands exponentially while satisfaction and quality decline, because the constraint of limited time used to force you to pick the right problem.
"Token Anxiety" is the new form of the programmer's inability to disconnect -- except instead of occasionally swirling on a hard bug, it is now a constant background hum of agents running, prompts waiting, and the nagging feeling that something could be running right now.
Vibe coding produces fast MVPs but terrible code quality; Prime found that every fix required more prompting in increasingly long cycles, and running multiple agents simultaneously led to constant context-switching and babysitting rather than focused work.
The most important filter that AI removes -- having to be selective about which ideas to pursue -- was never a bottleneck but a feature. Without it, developers end up with many crappy versions of many ideas and nothing actually shipped.
Prime's core advice: hard work is a tool, not an identity. One extra feature is not worth missing meaningful time with friends and family -- a lesson he learned the hard way during his startup years and is having to re-learn in the age of AI agents.
The Breakdown
Prime opens the video by warning viewers this one is going to be different -- more serious, more personal. He wants to talk about a growing trend he is seeing among developers: a perpetual, low-grade anxiety driven by AI agents and the vibe coding era. He calls it a weird, almost spiritual shift in how programmers relate to their work and their downtime.
To set the stage, he reflects on his earlier programming years. When he hit a particularly gnarly bug, he would step away from the keyboard, take a shower, and try to think about nothing. Inevitably, the solution would surface on its own -- a "I know kung fu" moment mid-shower. He describes this as a kind of superpower, but also acknowledges its dark side: the inability to fully disconnect. He would sit at dinner with his wife or hang out with friends, and part of his brain was always swirling on the problem in the background. Anyone who has programmed long enough, he says, knows that exact feeling.
Prime then pivots to an article he came across called "Token Anxiety," which describes life in present-day San Francisco. He contrasts this with his own experience arriving in the Bay Area in 2013. Back then, the energy was optimistic and raw -- everyone at parties was talking about what they were building, what their startup was doing, feeling like they were on the precipice of the next big thing. By the time he left in 2020, that energy had already faded somewhat. But what has replaced it now, he argues, is something far worse: not excitement, but desperation.
He reads passages from the article that paint a stark picture. A friend left a party at 9:30 on a Saturday -- not sick, not tired -- because he wanted to get back to his AI agents. Half the room was thinking the same thing. Young people in the Bay are going sober now, not for health reasons, but because they plan to go back to work after the party. He references the CEO of Y Combinator tweeting that he is giving up drinking because of Claude Code so he can "sling 10,000 lines of code a day." The old Bay Area optimism -- "there's so much opportunity, you have to come here" -- has inverted into a fear of falling behind: "I better keep working because if I don't, it's escaping."
The article's core point, which Prime agrees with, is that the anxiety is rational, and that is precisely why it sticks. Every week a new benchmark drops, a new model ships, context windows double. None of it reduces the pressure -- it multiplies it. You can do more now, and someone already is. One passage describes replacing Netflix with Claude Code, lying in bed thinking about what agents to spin up before falling asleep, feeling that reading a novel is indulgent and watching a movie without a laptop open is wasteful. Prime reads this and says plainly: "It just feels sad."
He then turns inward, admitting he has been deliberately immersing himself in the vibe coding world to understand it. He built several vibe-coded tools for his stream -- a Twitch chat controller, an OBS management app, a meme search tool. The MVP comes together fast, but every fix requires more prompting, more waiting, and the cycle is painfully long. Worse, because spinning up a new project is so easy, he found himself running three or four agents simultaneously, none of them producing truly good code, the work growing exponentially. He describes it as a Faustian bargain: you are building more than you have ever built, but you are not actually accomplishing the things you set out to accomplish. Ten thousand lines of code in a day, sure -- but none of it satisfying, none of it necessarily the right thing.
Prime identifies what he thinks is the deepest problem: before AI agents, you could not try out every idea, so you had to be selective. That constraint was actually a gift. Now you can spin up every half-formed notion, and each one spawns its own expanding universe of work. The programming was never the hard part -- solving the right problem was. He mentions people in the Bay spending $1,000 a day on tokens, jumping from thing to thing, productivity metrics through the roof, yet having literally nothing to show for it.
He closes with personal advice, drawing on a parallel from 2009-2010 when he was chasing the Silicon Valley startup dream, pulling all-nighters in NetBeans writing PHP, feeling anxious to leave friends and get back to coding. That pursuit strained his friendships and his marriage. His advice to viewers, and to himself: that one extra feature in your calendar app is not worth skipping out on good times with your friends. Hard work got him to where he is, but it is not who he is. He signs off acknowledging the video might not have been entirely coherent, but he needed to get it off his chest.