Hard Fork Podcast Summary

Episode: ‘A.I.-Washing’ Layoffs? + Why L.L.M.s Can’t Write Well + Tokenmaxxing
Hosts: Kevin Roose (The New York Times), Casey Newton (Platformer)
Guest: Jasmine Sun (Freelance Journalist)
Date: March 20, 2026

Episode Theme

This episode explores seismic shifts in the tech industry brought on by AI, focusing on three intertwined topics: whether recent tech layoffs are truly caused by AI (“A.I.-washing”), why large language models (LLMs) seem to hit a ceiling on good writing, and the rise of token “leaderboards” that track who is spending the most on AI tools within tech companies. The hosts, joined by Jasmine Sun, dig deep into the reality behind the headlines and examine the human and economic stakes of AI’s next wave.

1. ‘A.I.-Washing’ and Tech Layoffs

[02:00–17:55]

Key Points

Recent Wave of Layoffs: Tech giants such as Atlassian, Block, and reportedly Meta are cutting large portions of their workforce and citing AI-driven efficiency as partial justification.
AI as an Excuse (‘AI-Washing’): The term "AI-washing" is discussed: Are companies using AI as a convenient cover for cost-cutting, or is AI truly displacing jobs?
- Quote (Casey, 06:29): “I thought it was when a software engineer finally took a shower.”
- Quote (Kevin, 06:34): “Basically the thesis is like, these aren't really layoffs about AI. This is just sort of a convenient excuse that these companies are using.”
Economic Drivers: Company narratives often align to please markets; referencing AI can boost stock prices, reminiscent of past crypto hype.
Shift in Spending: Layoffs are not always reducing costs—they’re often redirecting budget from people to AI infrastructure and data centers.
- Quote (Kevin, 11:40): “They are plowing this money that they are going to save by laying off these thousands of people into the building of data centers and other AI infrastructure.”
Worker Anxiety and Labor Power:
- Employees wonder if using AI makes them more valuable or simply easier to automate away.
- Hostile layoffs have made workers more compliant; unionization may become a more pressing topic, marking a major shift from earlier tech labor relations.
- Quote (Casey, 17:26): “I cannot think of anything that would make Mark Zuckerberg more mad than a union of software engineers at Meta.”

Notable Moments

Jokes about “AI washing” paralleling “Crypto washing” and superficial market tricks.
Direct acknowledgment of personal ties and biases, e.g. Casey’s fiancé working at Anthropic.

2. Why LLMs Can’t Write Well

Guest: Jasmine Sun, Author of “The Human Skill That Eludes AI”
[19:46–44:59]

Key Points

LLMs Still Struggle With Truly Good Writing: Despite advances, LLM outputs tend to be formulaic or lack genuine literary flair, especially in creative or lived-experience writing.
- Quote (Jasmine, 21:55): “Most writing period is very bad. And so I think that language models are definitely better at writing and language than most humans are. But... why can't they write at a sort of literary, creative fiction level?”
Comparison Across Model Eras:
- Jasmine found earlier models (e.g. GPT-2, GPT-3) sometimes surprisingly stylistic, even weird, compared to sanitized modern models like ChatGPT 5.4, which are “optimized for blandness.”
- Quote (Jasmine, 23:12): “I found so much more compelling than ChatGPT today. It doesn't have any of the annoying ticks, it doesn't have the M dashes, the tripartite list.”
What Happened? RLHF and Post-Training:
- “Alignment” and reinforcement learning from human feedback (RLHF) have made LLMs safer and more useful but also boring.
- Human editors are often asked to judge “quality” using mismatched rubrics (e.g., limiting exclamation marks or grading fan fiction on factuality).
- Quote (Jasmine, 27:02): “The rubrics just didn't make any sense. He would be told things like, you have to grade them based on the number of exclamation marks… He got a bunch of fan fictions and he was supposed to grade them on their factuality.”
LLMs Lack Lived Experience:
- At their creative best, writing draws on specific, lived experience; LLMs only remix patterns from their data.
- Quote (Jasmine, 31:49): “They don't have lives. That means that all of the metaphors they choose, all of the words they choose, the examples they choose, they're just ungrounded.”
Objections and Debate:
- Would AI rapidly catch up with “true” writing if the incentive shifted?
- Is our resistance to AI-generated literature just bias—a “blind taste test” paradox?
- Quote (Kevin, 35:02): “Is it possible that the models have already become superhuman at writing, but the minute we learn that they are AI models generating text… we lose all interest in it just because of the source, not because of the quality?”
Human-AI Collaboration, Not Competition:
- Jasmine describes her own method for using Claude as a partner in editing—training its critique to her personal standards, not generic rubrics. This achieves better editorial guidance and self-improvement.
- Quote (Jasmine, 41:11): “I wanted to learn what do I aspire to be and where do I see myself falling short and where. What am I proud of, right?... we were able to co develop a rubric of... qualitative criteria.”

Notable Moments

Recalling Sam Altman’s hedging: promising AI super-skills but conceding, “maybe in the future ChatGPT will be able to write, quote, a real poet's okay poem” (Jasmine, 21:55).
Discussion of the “centaur” model: best work comes from human + AI collaboration.
Jasmine’s transparency about her own attempts to automate herself—or not—and a candid, practical segment on how writers use AI to up their game.

3. Tokenmaxxing: The AI Leaderboard Craze

[47:21–62:53]

Key Points

Token Leaderboards:
- New trend at AI-forward tech companies: tracking and ranking employees by number of AI “tokens” they consume—essentially measuring the atomic units of AI-generated work.
- Quote (Kevin, 47:45): “It’s a token frenzy out there, and the employees of these companies are competing among their colleagues... They want to be the people at their company who are using the most AI tokens.”
Why Track Tokens?
- Seen as a proxy for adoption of new AI tools and productivity.
- Top users (“billion-token club”) spend staggering amounts on tokens—sometimes more than their salary, with free use for in-house employees.
- Quote (Kevin, 51:39): “The top user of CLAUDE code, the top individual user of CLAUDE code as measured by Anthropic, spent more than $150,000 on tokens last month.”
Bad Incentives & Goodhart’s Law:
- Creates perverse incentives to “waste” tokens or start side hustles rather than do truly valuable work.
- Quote (Casey, 53:51): “That just seems like it would create the worst incentives. Right. There's this idea of Goodhart's law, right. Like when a measure becomes a target, it seem pieces to become a good measure.”
Broader Implications & Transfer to Other Industries:
- Creating leaderboards for AI use in other domains (e.g., marketing) may incentivize performative rather than productive behavior.
- Managers need to focus on actual value, not just metrics.
- Job negotiations at AI labs now consider “token budgets” as a perk or necessity for power users.

Notable Quotes

Casey, 58:58: “I have been struck at how this idea of the token leaderboard just represents a new incarnation of something that the software industry has been trying to figure out for a long time, which is how can I figure out if my software engineers are productive?”
Kevin, 61:17: “There will be people who are token maxing who are way more productive than their colleagues and doing way more projects way more quickly. I think there will be other people whose managers look at their, like, token budgets and see, say you spent this many tokens on what? And we'll have to have some hard conversations.”

Additional Notable Quotes and Timestamps

On ‘A.I.-Washing’
- Kevin, [09:57]: “There’s sort of this narrative power around AI where if you seem like a company that is investing heavily in the AI tools and the AI way of working, your investors say, oh, that company is really forward looking.”
- Casey, [10:08]: “It turns out that the public markets actually can just be tricked that easily.”
On Creative AI
- Jasmine, [31:49]: “...their writing has stakes. It comes from an emotional place. And the fact that LLMs... don't have lives. That means that all of the metaphors they choose... they're just ungrounded.”
On Editing with AI
- Jasmine, [41:11]: “We're co developing these qualitative criteria and then I split it into phases...I put this all in a cloud project. I said, your job is to evaluate my drafts based on this criteria, but not to do the writing for me and to make sure to prompt out of me what I can do better.”

Episode Flow and Tone

Style: The episode blends journalistic curiosity, playful banter, and skepticism. Kevin and Casey riff on headlines and hold nuanced discussions—mixing skepticism, optimism, and humor.
Language: Conversational, at times irreverent ("Jay Z washing," "software engineer finally took a shower"), with sharp, accessible explanations of technical and economic concepts.

Conclusion

This episode is a deep dive into the murky realities behind AI’s market hype: major tech layoffs justified (maybe opportunistically) under the AI banner, the struggle of even “superhuman” LLMs to write with real voice and soul, and the wonky new culture of “tokenmaxxing” that’s reshaping incentives and office politics. The conversation with Jasmine Sun is particularly rich for anyone interested in how AI is (and isn’t) changing creative work.

For Further Exploration

Hard Fork Podcast Summary

Episode Theme

1. ‘A.I.-Washing’ and Tech Layoffs

[02:00–17:55]

Key Points

Recent Wave of Layoffs: Tech giants such as Atlassian, Block, and reportedly Meta are cutting large portions of their workforce and citing AI-driven efficiency as partial justification.
AI as an Excuse (‘AI-Washing’): The term "AI-washing" is discussed: Are companies using AI as a convenient cover for cost-cutting, or is AI truly displacing jobs?
- Quote (Casey, 06:29): “I thought it was when a software engineer finally took a shower.”
- Quote (Kevin, 06:34): “Basically the thesis is like, these aren't really layoffs about AI. This is just sort of a convenient excuse that these companies are using.”
Economic Drivers: Company narratives often align to please markets; referencing AI can boost stock prices, reminiscent of past crypto hype.
Shift in Spending: Layoffs are not always reducing costs—they’re often redirecting budget from people to AI infrastructure and data centers.
- Quote (Kevin, 11:40): “They are plowing this money that they are going to save by laying off these thousands of people into the building of data centers and other AI infrastructure.”
Worker Anxiety and Labor Power:
- Employees wonder if using AI makes them more valuable or simply easier to automate away.
- Hostile layoffs have made workers more compliant; unionization may become a more pressing topic, marking a major shift from earlier tech labor relations.
- Quote (Casey, 17:26): “I cannot think of anything that would make Mark Zuckerberg more mad than a union of software engineers at Meta.”

Notable Moments

Jokes about “AI washing” paralleling “Crypto washing” and superficial market tricks.
Direct acknowledgment of personal ties and biases, e.g. Casey’s fiancé working at Anthropic.

2. Why LLMs Can’t Write Well

Guest: Jasmine Sun, Author of “The Human Skill That Eludes AI”
[19:46–44:59]

Key Points

LLMs Still Struggle With Truly Good Writing: Despite advances, LLM outputs tend to be formulaic or lack genuine literary flair, especially in creative or lived-experience writing.
- Quote (Jasmine, 21:55): “Most writing period is very bad. And so I think that language models are definitely better at writing and language than most humans are. But... why can't they write at a sort of literary, creative fiction level?”
Comparison Across Model Eras:
- Jasmine found earlier models (e.g. GPT-2, GPT-3) sometimes surprisingly stylistic, even weird, compared to sanitized modern models like ChatGPT 5.4, which are “optimized for blandness.”
- Quote (Jasmine, 23:12): “I found so much more compelling than ChatGPT today. It doesn't have any of the annoying ticks, it doesn't have the M dashes, the tripartite list.”
What Happened? RLHF and Post-Training:
- “Alignment” and reinforcement learning from human feedback (RLHF) have made LLMs safer and more useful but also boring.
- Human editors are often asked to judge “quality” using mismatched rubrics (e.g., limiting exclamation marks or grading fan fiction on factuality).
- Quote (Jasmine, 27:02): “The rubrics just didn't make any sense. He would be told things like, you have to grade them based on the number of exclamation marks… He got a bunch of fan fictions and he was supposed to grade them on their factuality.”
LLMs Lack Lived Experience:
- At their creative best, writing draws on specific, lived experience; LLMs only remix patterns from their data.
- Quote (Jasmine, 31:49): “They don't have lives. That means that all of the metaphors they choose, all of the words they choose, the examples they choose, they're just ungrounded.”
Objections and Debate:
- Would AI rapidly catch up with “true” writing if the incentive shifted?
- Is our resistance to AI-generated literature just bias—a “blind taste test” paradox?
- Quote (Kevin, 35:02): “Is it possible that the models have already become superhuman at writing, but the minute we learn that they are AI models generating text… we lose all interest in it just because of the source, not because of the quality?”
Human-AI Collaboration, Not Competition:
- Jasmine describes her own method for using Claude as a partner in editing—training its critique to her personal standards, not generic rubrics. This achieves better editorial guidance and self-improvement.
- Quote (Jasmine, 41:11): “I wanted to learn what do I aspire to be and where do I see myself falling short and where. What am I proud of, right?... we were able to co develop a rubric of... qualitative criteria.”

Notable Moments

Recalling Sam Altman’s hedging: promising AI super-skills but conceding, “maybe in the future ChatGPT will be able to write, quote, a real poet's okay poem” (Jasmine, 21:55).
Discussion of the “centaur” model: best work comes from human + AI collaboration.
Jasmine’s transparency about her own attempts to automate herself—or not—and a candid, practical segment on how writers use AI to up their game.

3. Tokenmaxxing: The AI Leaderboard Craze

[47:21–62:53]

Key Points

Token Leaderboards:
- New trend at AI-forward tech companies: tracking and ranking employees by number of AI “tokens” they consume—essentially measuring the atomic units of AI-generated work.
- Quote (Kevin, 47:45): “It’s a token frenzy out there, and the employees of these companies are competing among their colleagues... They want to be the people at their company who are using the most AI tokens.”
Why Track Tokens?
- Seen as a proxy for adoption of new AI tools and productivity.
- Top users (“billion-token club”) spend staggering amounts on tokens—sometimes more than their salary, with free use for in-house employees.
- Quote (Kevin, 51:39): “The top user of CLAUDE code, the top individual user of CLAUDE code as measured by Anthropic, spent more than $150,000 on tokens last month.”
Bad Incentives & Goodhart’s Law:
- Creates perverse incentives to “waste” tokens or start side hustles rather than do truly valuable work.
- Quote (Casey, 53:51): “That just seems like it would create the worst incentives. Right. There's this idea of Goodhart's law, right. Like when a measure becomes a target, it seem pieces to become a good measure.”
Broader Implications & Transfer to Other Industries:
- Creating leaderboards for AI use in other domains (e.g., marketing) may incentivize performative rather than productive behavior.
- Managers need to focus on actual value, not just metrics.
- Job negotiations at AI labs now consider “token budgets” as a perk or necessity for power users.

Notable Quotes

Casey, 58:58: “I have been struck at how this idea of the token leaderboard just represents a new incarnation of something that the software industry has been trying to figure out for a long time, which is how can I figure out if my software engineers are productive?”
Kevin, 61:17: “There will be people who are token maxing who are way more productive than their colleagues and doing way more projects way more quickly. I think there will be other people whose managers look at their, like, token budgets and see, say you spent this many tokens on what? And we'll have to have some hard conversations.”

Additional Notable Quotes and Timestamps

On ‘A.I.-Washing’
- Kevin, [09:57]: “There’s sort of this narrative power around AI where if you seem like a company that is investing heavily in the AI tools and the AI way of working, your investors say, oh, that company is really forward looking.”
- Casey, [10:08]: “It turns out that the public markets actually can just be tricked that easily.”
On Creative AI
- Jasmine, [31:49]: “...their writing has stakes. It comes from an emotional place. And the fact that LLMs... don't have lives. That means that all of the metaphors they choose... they're just ungrounded.”
On Editing with AI
- Jasmine, [41:11]: “We're co developing these qualitative criteria and then I split it into phases...I put this all in a cloud project. I said, your job is to evaluate my drafts based on this criteria, but not to do the writing for me and to make sure to prompt out of me what I can do better.”

Episode Flow and Tone

Style: The episode blends journalistic curiosity, playful banter, and skepticism. Kevin and Casey riff on headlines and hold nuanced discussions—mixing skepticism, optimism, and humor.
Language: Conversational, at times irreverent ("Jay Z washing," "software engineer finally took a shower"), with sharp, accessible explanations of technical and economic concepts.