The AI Productivity Trap — And What Comes After
A few weeks ago, Peter Steinberger — iOS developer turned prolific AI builder — gave a talk titled “Just One More Prompt” at a Claude Code Anonymous meetup in London. Yes, that’s a real thing.
His confession was honest and familiar: since discovering agentic AI, he can’t stop building. He works through the night. He texts friends at 4am and they’re also awake. He calls them “the Black Eye Club.” He built a tool so he could talk to Claude from his phone — then realized he’d essentially built better access to his drug.
He’s not alone. This is becoming an industry-wide pattern.
The CEOs Who Think This Is a Feature
Here’s what leadership looks like in 2025-2026:
Cognition CEO Scott Wu:
“Cognition has an extreme performance culture. We routinely are at the office through the weekend and do some of our best work late into the night. Many of us literally live where we work.”
Kennan Davison, Icon (AI Marketing):
“Why do 6 [days of work] when you can do 7”
Google co-founder Sergey Brin:
“60 hours a week is the sweet spot of productivity”
Read those again. These aren’t junior developers bragging on Twitter. These are the people shaping the industry. And their message is clear: AI makes you more productive, so you should work more.
That framing is the trap.
The Paradox Nobody Talks About
AI was supposed to save us time. Instead, the most engaged AI users — founders, developers, operators — report working more hours than before. More FOMO. More ideas. More “just one more prompt” at 2am.
The technology that was supposed to free us is creating a new kind of addiction. Not to social media. Not to entertainment. To building. To the intoxicating feeling that you can finally do everything you ever dreamed of — if you just don’t stop.
Peter describes it perfectly: “One week in AI feels like a month in the real world.”
And here’s what’s wild — he knows it’s unsustainable. He literally says “I realize it’s time to slow down.” His solution? Adding a session timer to Claude’s status line. A tiny reminder that time is flying while you’re in the flow.
A timer. That’s the best we’ve got.
The Real Problem: Humans Became the Operating System
Every productivity tool in the last 20 years increased our capability. Email. Slack. Notion. Asana. AI. Each one gave us more power. And each one offloaded the coordination back to us.
We became the memory layer (remembering what was decided, what was promised), the integration layer (connecting tools, channels, people), the supervision layer (making sure things actually happened), and the decision layer (processing everything, constantly).
The tools got smarter. The human got more loaded.
AI amplified this pattern instead of breaking it. GPT doesn’t reduce your cognitive load — it increases your output capacity while you remain the bottleneck for everything else. You can write faster, code faster, research faster. But you’re still the one holding all the threads.
That’s not productivity. That’s a faster hamster wheel.
The Measurement Problem
When Sergey Brin says “60 hours is the sweet spot,” he’s measuring the wrong thing. Hours are a proxy metric from the industrial era. They measure presence, not impact. They measure input, not outcome.
When Scott Wu celebrates sleeping at the office, he’s not describing innovation — he’s describing a system that hasn’t figured out how to separate human judgment from human labor.
And when Peter Steinberger builds a tool to access Claude from his phone at 4am, the question isn’t “how do we make that faster?” The question is “why is a brilliant builder unable to stop?”
The answer: because the system has no off switch. There’s no layer between the human and the work that says “this can wait” or “this is already handled” or “you’ve made enough decisions today.”
What “After” Looks Like
There’s a different way to think about this. Not AI that helps you do more. AI that does things so you don’t have to.
Not a copilot. Not an assistant that waits for your prompt. A system that tracks your commitments so you don’t carry them in your head, closes loops so you stop worrying about what fell through, protects your focus so ideas don’t hijack your week, operates across your channels so you’re not the integration layer, and knows when to act and when to stay silent.
The shift isn’t about using AI more. It’s about using it as a production line — where each stage does its job autonomously and hands off to the next. From AI as amplifier to AI as infrastructure. From “help me do more” to “handle this so I can be present.”
The real solution isn’t discipline — it’s building a system that maintains itself. One that catches you before the hamster wheel spins out.
Present for dinner. Present for the conversation. Present for the walk. Present for the thinking that only happens when you’re not reacting.
The Anti-Productivity Metric
What if we measured success differently?
Not: “How many hours did I work?” But: “How many decisions resolved without me?”
Not: “How much did I ship?” But: “How calm did I feel this week?”
Not: “How fast did I respond?” But: “How few things required my attention at all?”
This isn’t laziness. It’s architecture. The most effective systems aren’t the ones where the operator works the hardest. They’re the ones where the operator intervenes the least.
A well-run factory doesn’t need the CEO on the floor 60 hours a week. A well-built AI layer shouldn’t need you awake at 4am.
The Choice
Peter’s story is a love letter to building. And building is beautiful. The problem isn’t the passion — it’s the absence of a system that catches you.
The CEOs celebrating 80-hour weeks aren’t wrong about the opportunity. They’re wrong about the response. The correct response to “AI can do incredible things” isn’t “so I should work more.” It’s “so I should design a system where the incredible things happen — and I get my life back.”
That’s the transition we’re in. From AI-assisted humans to AI-sustained systems. From the human as operating system to the human as the one who gets to live.
Peter added a timer. That’s a start.
But what if the system didn’t need a timer — because it already knew when to stop?
I’m Eliran — thinking about what happens when AI stops being a tool and starts being infrastructure. More at elirank1.github.io/blog.
Peter’s original talk: Just One More Prompt — worth reading in full.
