Why I'm Skeptical of the AI Vampire 3-to-4-Hour Workday — Raising Human Intensity Every Day Instead

I read Steve Yegge’s The AI Vampire. His central claim is that AI coding agents, driven by tools like Claude Code, really do deliver a 10x productivity boost, but the humans who wield them are being drained Colin Robinson style. His conclusion is crisp: 3 to 4 hours is the new sweet spot for a workday.

The passage I want to push back on is this one.

I’m convinced that 3 to 4 hours is going to be the sweet spot for the new workday. Give people unlimited tokens, but only let people stare at reports and make decisions for short stretches. Assume that exhaustion is the norm. Building things with AI takes a lot of human energy.

I don’t disagree with the vampire effect itself. The fatigue after running an agent hard for a long stretch is qualitatively different from the fatigue after writing the same code by hand. Decision density is higher, and every decision reshuffles the assumptions you were holding in your head a minute earlier. What I’m skeptical of is the derived conclusion (that 3 to 4 hours becomes the sweet spot). This post lays out two reasons for that skepticism.

Restating Yegge’s Argument

Let me summarize the argument in my own words first.

At the level of Opus 4.5 / Opus 4.6 plus Claude Code, AI coding produces something close to a 10x productivity boost in practice
The human side of the loop, however, burns energy on decisions, reviews, and staring at reports
As a result, an all-in AI workflow cannot be sustained for 8 hours a day
Whether the individual or the company captures that 10x value cannot be at either extreme (Scenario A or Scenario B); the dial has to settle in the middle
The proposed middle setting is a 3-to-4-hour workday

I follow the reasoning that Scenario A (the employee runs 10x for 8 hours until burnout) and Scenario B (the employee runs 1 hour a day and the company folds) both fail. The dial has to sit in the middle. What I’m not convinced by is that “the middle” equals 3 to 4 hours. The two sections below are where I part ways.

Skepticism 1: You Need Daily Acclimatization to How Fast AI and Agents Are Changing

The first reason is the pace at which the AI tooling stack itself is evolving.

Models, agent CLIs, agent frameworks, MCP servers, and the surrounding tools all get refreshed on something like a weekly cadence. Best practices that were state of the art a few weeks ago flip when a new model or a new agent behavior lands. Codex CLI, Cursor, and their peers are moving targets in themselves.

If you compress your working day into 3 to 4 hours during this phase, you fall behind the rate of change in a structural way. Really learning a new model requires putting it against real tasks, watching both its failures and successes, and rewiring your existing workflow around it. That kind of learning does not come from reading release notes. It scales with hands-on time.

Someone who is exposed to AI and agents every day at elevated intensity and someone who touches them only inside a 3-to-4-hour window will diverge substantially in half a year. The gap is not just familiarity with the tools. It is the harder-earned judgment about where an agent will run off the rails, how to set stop conditions, what prompt granularity works, and how to grow assets like AGENTS.md or CLAUDE.md over time.

When the environment shifts abruptly, that gap is not something you can close in a week. Capping your working hours protects your rhythm, but the cost is that you are silently taking on the risk of not being able to catch up when the ground moves. That is why I think the safer play right now is to raise intensity a little every day, so your body stays acclimated to the pace at which AI and agents are changing.

Skepticism 2: A High-Intensity Human Can Run AI at Full Throttle Over Long Hours

The second reason is about where the market equilibrium actually lands.

Yegge argues that the industry should turn the dial to 3 to 4 hours. On an individual level, that is a decision anyone can make. On a company level, it is operationally feasible. But at the level of a global market, there will always be players who happen to have high baseline intensity and can run AI at full throttle over long hours. Given the same models and the same token budget, a 3-to-4-hour setup cannot match their raw output.

The Scenario A / Scenario B framing works as a within-company contrast between the individual and the employer. It stops working once you widen the frame to the whole market, because other kinds of players enter.

Individuals with high baseline intensity whose body, environment, and life setup permit long-hour operation
Startups and owner-operated companies that don’t face fixed working-hour constraints
Distributed teams in regions with different labor-hour norms, or teams that use time zones to keep the workflow running continuously

These groups can, with reasonable safeguards against vampire fatigue, operate at something other than 3 to 4 hours per day. If they choose an 8-hour or 10-hour setting, a 3-to-4-hour cluster using the same AI simply cannot catch up on raw output.

That is why I think the “average settles at 3 to 4 hours” reading underestimates the market’s asymmetry. The pressure is toward the high-intensity side of the dial, not the low. The median setting will land higher than Yegge’s estimate.

The Vampire Effect Is Still Real

None of the above means the vampire effect can be ignored. The fatigue after a long agent-driven stretch is real, and decision quality drops when you push it too far. Choosing to raise intensity trades off stamina, focus, and margin in daily life, so it is not something you can commit to by simply declaring “I’ll run AI for 10 hours a day from tomorrow.”

The working assumptions I hold right now are these.

Treat fatigue as a variable to design for daily, not as an abnormal state to avoid
Deliberately separate short high-intensity sessions from lower-intensity work
Review the ceiling of what agents can take off your plate every week; do not mechanically extend human continuous-operation time
Book sleep, physical activity, and off-screen time up front, as investments that make higher intensity possible
Refresh the intensity plan and workflow design together, timed to model / agent generation transitions

Rather than starting from a 3-to-4-hour ceiling, start by designing a workflow that does not break at higher intensity. The order matters.

Summary

Steve Yegge’s The AI Vampire correctly names the fatigue that comes with a 10x productivity boost from AI
I disagree with the conclusion that 3 to 4 hours becomes the new sweet spot, for two reasons
1. AI and agents are evolving fast enough that daily exposure is necessary to stay adaptable
2. As long as high-intensity humans can run AI at full throttle over long hours, the market equilibrium tilts toward the high end of the dial, not the low
The vampire effect is real, so intensity has to be paired with a daily design of fatigue, sleep, and off-screen time
The right framing is not to cap the workday at 3 to 4 hours, but to design a workflow that survives higher intensity

That’s all from pushing back on the 3-to-4-hour workday reading of the AI vampire and proposing daily acclimatization to AI and agents instead, from the Gemba.