The AI Vampire

Source: Steve Yegge Date: 2026-02-11 URL: https://steve-yegge.medium.com/the-ai-vampire-eda6e4f07163?source=rss-c1ec701babb7------2

Summary

Yegge argues that AI tools create a productivity paradox: genuine output gains become an “energy vampire” because companies capture the value while workers bear the exhaustion. Early adopters set unrealistic standards; startups perpetuate burnout culture chasing extraction; enterprises then demand the same intensity. His prescription is workers controlling the denominator — hours worked — capping at 3–4 hours of high-intensity agentic coding per day. He applies this to himself: a deliberate retreat from overcommitment.

Implications

The human-cost thread. This is the clearest statement from a prominent agent practitioner that current agentic workflows are unsustainable at full-throttle. The “nap strike” from the birthday blog is the symptom; this post is the diagnosis. Any orchestration tool that reduces decision load on the human — better async delegation, clearer task decomposition, less context-switching — has a direct argument here.

Pressure on enterprise AI adoption narrative. The dominant enterprise pitch for coding agents is productivity gain. Yegge is arguing that gain is real but the extraction model is broken — it transfers value upward while distributing fatigue downward. That’s a union argument, not a product argument. Worth watching if this framing gets picked up by labor discourse around AI-augmented knowledge work.

Implication for tooling design. If 3–4 hours is the sustainable ceiling for high-quality agentic work, tooling that makes those hours more leverage-efficient wins over tooling that optimizes for raw throughput. Async handoff, offline agent continuation, and better interrupt/resume become differentiators.

Watch: whether sustainable-AI-work becomes an explicit design criterion in any major agent product; whether Yegge’s own Gas Town/Beads work starts incorporating explicit cognitive-load controls.