Ed Zitron loses his mind annotating an AI doomer macro memo

Scope of the annotation and memo

• The annotated document is a critique of a highly bullish “AI doomer” macro memo that briefly moved markets.
• Commenters say the memo blends kernels of truth (especially about labor and macro risk) with speculative “fantasy doomer porn” about ultra‑automation.
• Some see Zitron’s annotation as a needed takedown of overblown rhetoric; others find his tone juvenile and one‑sided.

Capabilities and limits of LLMs / coding agents

• Ongoing dispute over whether coding agents are “successful”:
  • Critics: generated code is often wrong, encourages low‑skill “slop,” and increases review burden on better developers.
  • Supporters: when used well, tools like Claude Code materially speed up development; many companies are already building or replacing substantial internal systems with LLM help.
• Debate over hallucinations:
  • Some argue they are fundamentally unsolvable for token‑predictors.
  • Others counter that current models are already “reliable enough” for many tasks and measurable hallucination rates have sharply improved.

Economic viability and costs

• One camp says Zitron is wrong because:
  • Inference prices per capability have dropped 20–900x over time (citing datasets like Epoch / Artificial Analysis).
  • Open and Chinese models report very high theoretical margins and show similar cost declines.
• A skeptical camp counters:
  • Public “price per token” says nothing about true unit costs or whether prices are massively subsidized.
  • Training and hardware CAPEX (chips, multi‑GW datacenters, trillions in projected spend) is the real risk; demand forecasts can easily be wrong and are non‑hedgeable.
  • Rapid price declines imply brutal asset depreciation and heighten business risk rather than reducing it.

Work, burnout, and quality

• Several developers report:
  • Forced “AI‑first” policies, machine‑reviewed PRs, and colleagues shipping code they don’t understand.
  • Fear that the only way these investments make sense is if software production becomes largely automated, with grim implications for careers.
• Others say AI already enables leaner teams and internal build‑vs‑buy shifts (e.g., reconsidering Salesforce‑like subscriptions), which may pressure SaaS prices.

Views on Zitron and overall tone

• Some see him as a necessary counterweight who digs into numbers and punctures hype.
• Others call him irrational, grifting, or technologically illiterate, and argue his constant dismissal of AI’s usefulness is misleading and harmful to anxious engineers.
• A few commenters are exhausted by both extremes—doomer macro memos and sneering counter‑polemics—and ask for more measured, less hysterical discussion.