AI Layoffs, Compute Costs & Agents | Naveen Rao & Alex Finn on This Week in AI Episode 16

TL;DR

• 1

  A lot of AI cost inflation is fake demand: Naveen Rao says roughly 20% to 30% of token spend may come from “token maxing,” where companies gamify usage with simple metrics and leaderboards instead of tracking meaningful output.

• 2

  AI is a force multiplier for skilled people, not a replacement for them: Alex Finn says AI already makes him “a thousand times faster,” while Rao argues it still cannot reliably replace a senior engineer making decisions about security, extensibility, and production quality.

• 3

  Frontier models keep pricing power because teams use them to reduce risk: Rao says he probably only needs the best model 5% of the time, but because it is hard to know in advance which tasks need it, he often defaults to the expensive option anyway.

• 4

  The next big compute constraint is energy, not just chips or memory: Rao says energy was about 7% to 10% of A100 total cost of ownership, is nearing 40% for current Nvidia generations, and could exceed 50% next generation, which is why Unconventional AI is chasing 100x to 1000x power efficiency gains.

• 5

  Small AI-native teams still want more engineers, not fewer: At Unconventional AI, a finance leader built internal recruiting agents with vibe coding, which improved coverage and surfaced missed candidates, but then created even more worthwhile engineering work that required hiring developers.

• 6

  AI has a political and PR problem, not just a product problem: Finn and Rao both argue that layoffs are often being blamed on AI to cover for years of overhiring, while doomer messaging and local fears about data centers are turning AI into a public boogeyman instead of a visible source of shared benefit.