Hard Fork

‘A.I.-Washing’ Layoffs? + Why L.L.M.s Can’t Write Well + Tokenmaxxing

Hosted by Kevin Roose and Casey Newton
20 Mar 2026 8 min read 45m
TL;DR

Companies are increasingly using AI as a convenient justification for layoffs, but the economic reality is more nuanced—many layoffs stem from over-hiring and poor management rather than genuine AI displacement. The episode examines how AI-washing obscures the real structural problems in tech.

Key Moments
[No transcript — approximate]
“[No transcript — approximate] Companies are using AI as a smokescreen for layoffs that were driven by other factors like overexpansion and poor planning”
Main thesis of the episode discussing corporate communication around recent tech layoffs
[No transcript — approximate]
“[No transcript — approximate] The narrative of 'AI made our jobs obsolete' is easier to sell to investors than 'we hired too many people'”
Discussion of how companies frame workforce reductions to stakeholders
[No transcript — approximate]
“[No transcript — approximate] LLMs struggle with long-form coherent writing because they optimize for next-token prediction rather than narrative structure”
Segment on technical limitations of large language models in content creation
[No transcript — approximate]
“[No transcript — approximate] 'Tokenmaxxing' is the new arms race where companies compete on model scale rather than actual performance improvements”
Discussion of trends in AI development and diminishing returns on larger models
[No transcript — approximate]
“[No transcript — approximate] We're seeing a decoupling between AI hype and actual business value creation”
Closing analysis on the gap between AI expectations and reality in tech sector
About the show
›
Hard Fork is the New York Times' flagship podcast covering AI, big tech, and internet culture. Hosts Kevin Roose and Casey Newton break down the week's most important tech stories with sharp analysis and expert guests. The show explores how technology is reshaping business, society, and daily life.
Takeaways
1
AI-washing masks real corporate dysfunction Tech companies are leveraging AI anxiety to justify layoffs that actually stem from over-hiring cycles and management failures. This narrative shift lets leadership avoid accountability for poor planning while capitalizing on investor fears about automation. The result is a distorted public conversation about AI's actual impact on employment.
2
LLMs hit fundamental architectural limits Current large language models struggle with long-form writing because they predict tokens sequentially rather than planning narrative structure. This technical limitation means AI won't replace human writers for complex, coherent content anytime soon. Understanding these constraints is crucial for realistic product planning around AI capabilities.
3
Tokenmaxxing creates diminishing returns The AI industry is locked in a scale arms race where companies obsess over model size rather than actual performance gains or business applications. This approach wastes resources and masks slower progress on real AI problems. Product teams should focus on solving specific use cases rather than chasing headline-grabbing parameter counts.