+ +

Your agent swarm is running 24/7. You've got bots scraping Reddit at 6 AM, a resume updater, a competitor tracker, and a networking assistant that prompts you to follow up with people you forgot you met. You used to pay a human assistant in the Philippines. Now you just pay for tokens.

+ +

And somehow, you're more anxious than ever.

+ +

This isn't hypothetical. It's what's happening in Silicon Valley right now, and it's spreading. John Huang, a tech veteran who runs a startup investment network, has eight agents running simultaneously. Kevin Staight, a sales strategy lead, feels his "whole company should shut down and focus on this for about a month." Nikunj Kothari, a VC at FPV Ventures, has a name for what he sees everywhere in San Francisco:

+ +

Token anxiety.

+ +

Kothari describes it like this: "Everybody has this feeling of like, 'Hey, time is the only thing that matters. And in that given unit of time, which we don't get back, how can I have AI do a lot more for me than the next person?'"

+ +

He sees people checking their agents during parties. At bars. Walking through Dolores Park with "the warm glow of the light" from laptops slightly open. He himself has replaced his evening Netflix time with Claude Code, dreaming up new tasks to automate. He has two young kids and admits: "I felt guilty because I'd be like, 'Oh, go to bed quickly so I can get back to my computer.'"

+ +

The agents were supposed to free us. Instead, they've given us a new way to work during every waking hour — and a new reason to feel inadequate.

+ +

The Slot Machine Problem

+ +

Axios recently reported that AI agents "operate like slot machines" for power users. They produce just enough wins to keep you hooked — a clever automation here, a solved problem there — but the unpredictability means you keep pulling the lever. The wins are real, but so is the burnout.

+ +

The Allen Institute's Tim Dettmers notes that peak productivity now comes from "working with as many agents as possible in parallel," which requires near-constant context switching. Humans, as it turns out, aren't great at that.

+ +

We're building tools that demand exactly the cognitive pattern we're worst at.

+ +

The Junior Dev Canary

+ +

Stanford's 2026 AI Index dropped a staggering number: employment among software developers aged 22–25 has dropped nearly 20% since 2024, even as older developers' headcount grows. The same pattern appears in customer service and other AI-exposed roles.

+ +

We're not just automating tasks. We're automating the entry-level jobs that used to be the on-ramp to entire careers. The mentors who would have trained those junior devs are busy managing their agent swarms. The code reviews that would have taught them are being done by Claude.

+ +

Eric Weber, a former data/AI lead at Grammarly (now Superhuman), puts it plainly: "The change has happened so quickly that I think it is disorienting for people. People are like, 'Am I good at what I do? What does it mean to be good at what I do?'"

+ +

The industry, he says, is in a collective identity crisis.

+ +

Tokenmaxxing: When Companies Weaponize Your Anxiety

+ +

The New York Times recently reported that some of the biggest tech companies are "tokenmaxxing" — pushing employees to burn through as many AI credits as possible to churn out features at an ever-faster pace.

+ +

Think about that word for a second. It's not "being productive." It's not "leveraging AI." It's tokenmaxxing — a term that sounds like it fell out of a crypto bro's group chat. It treats AI usage as a metric to optimize, not a tool to wield.

+ +

When your company measures your value in tokens burned, you stop asking whether you're building the right thing. You just keep burning tokens.

+ +

What We're Not Asking

+ +

Weber's observation cuts to the core: "What is causing a lot of stress for people is that doing more doesn't necessarily create more leverage or impact. You're just doing more stuff. And so we don't spend a lot of time on the question: what should you actually be doing?"

+ +

That's the question the agent revolution hasn't answered. We've automated the how without examining the why. We've built systems that maximize output while ignoring outcomes.

+ +

A month ago, Weber stepped back from his full-time executive role. He still uses agents, but he's spending more time thinking, writing, and talking to people. He's asking the question that the token economy doesn't want you to ask.

+ +

Three Questions Before You Open That Laptop in the Park

+ +

Token anxiety isn't inevitable. It's a design choice — in our tools, in our companies, and in our own habits. Here are three questions worth asking before you start your next agent run:

+ +

1. What am I actually trying to accomplish? Not "what can I automate?" — what actually matters? If you can't articulate the outcome, the agent can't help you reach it. It can only make you busier.

+ +

2. Is this replacing work or replacing thinking? Agents should handle the work you understand well enough to delegate. They shouldn't replace the thinking that determines what work is worth doing. The moment you outsource judgment, you've stopped leading and started managing tokens.

+ +

3. Who am I not talking to right now? Every hour spent optimizing agent workflows is an hour not spent with your team, your family, your own thoughts. Weber stepped back to talk to people. Kothari catches himself rushing his kids to bed. The agent army doesn't need you to command it constantly. The people around you do.

+ +
+ +

The promise of agentic AI wasn't that we'd work more. It was that we'd work on what matters.

+ +

We're not there yet. But the path isn't to throw away the agents. It's to remember who's supposed to be in charge.

+ +