The New Problems AI Is Creating

Productivity Gains From AI Are Uneven
- AI productivity gains are jagged, with some tasks seeing immediate huge lifts while others resist change.
- Nathaniel explains organizations face integration work and oversight that often erases short-term time savings. (Time 0:03:12)
AI Is A Costly Operating Expense
- AI is not nearly free: operating costs scale with token consumption and frontier models use more tokens.
- Nathaniel notes firms exhausted budgets quickly and now implement token budgets, caps, and governance. (Time 0:06:11)
Allocate Model Power Deliberately
- Manage AI like capital allocation by matching model power to task and people to appropriate access levels.
- Nathaniel describes organizations building architectures of different models and budget pathways for higher access. (Time 0:08:32)
Clay’s Four Principle AI Writing Policy
- Clay instituted an AI writing policy focused on accountability, writing-as-thinking, reader respect, and brevity.
- Varun Anand’s four principles require authors to own ideas, avoid lazy long AI-generated docs, and sometimes just share prompts. (Time 0:11:41)
Combine Broad Access With Structured Experiments
- Give broad access to AI but pair it with structured experiments and top-down strategy to surface high-value use cases.
- Nathaniel cites OpenAI CFO Sarah Fryer’s zero-day close and continuous forecasting ambitions as examples. (Time 0:17:14)
Score AI By Value Per Unit Of Intelligence
- Measure AI by value per unit of intelligence, not seats or tokens; track cost including review and rework.
- Nathaniel relays Fryer’s four diagnostic questions about whether AI completed meaningful work and at what total cost. (Time 0:20:15)
Pick Lighthouse Teams And 10x Investment
- Avoid token-maxing or token-minimizing; instead pick lighthouse teams and 10x their investment to generate reproducible transformation.
- Nathaniel shares Section CEO Greg Shove’s guidance to prevent year-one stalls by getting specific on blockers. (Time 0:21:14)
Distributed De‑skilling Threatens Workforce Judgment
- A widespread risk is distributed de-skilling where judgment and problem framing erode as AI handles grunt tasks.
- Nathaniel cites BCG research: half of leaders already see skills softening and two-thirds expect a threat in 3–5 years. (Time 0:23:54)
Who Will Check AI When Experts Disappear
- The tragedy of the cognitive commons warns that removing junior grunt roles strips the pipeline that creates expert supervisors.
- Nathaniel cites Zara Zhang’s summary: AI consumes grunt work while expertise generation dries up, risking future inability to check AI. (Time 0:26:53)