Overview of Boss Class — Season 3, Episode 1: “Fat Layer of Humans”
This episode, hosted by Andrew Palmer of The Economist, launches Season 3 of Boss Class, which focuses on generative AI at work. Palmer mixes first‑person experiments (including a digital voice‑clone of himself) with interviews of people at different points on the AI frontier to explore where generative AI already helps, where it disappoints, and what managers and employees should do next.
Episode structure and key guests
- Opening: context for Season 3 (theme: generative AI at work) and Palmer’s voice‑clone demo (built by colleague Ruth Berry).
- Interview: Tom Blomfield (former Monzo co‑founder, partner at Y Combinator) — argues AI will dramatically thin the “fat layer of humans.”
- Interview: Ludwig Siegeler (The Economist tech editor) — introduces the idea of the “jagged frontier” (AI strong in some tasks, weak in others) and counsel to expect confusion.
- Interview: Ethan Mollick (Wharton professor, author of Co‑Intelligence) — practical guidance on when to use AI, organizational redesign, and experimentation.
- First‑person diary: Palmer experiments with various AI tools for research and writing; runs a blind test comparing his own comic column to one generated by an AI (Claude).
- Wrap and teaser: lessons learned and preview of later episodes that go deeper into AI deployments.
Main themes and insights
- The jagged frontier: AI capability is uneven — excellent at some narrowly defined tasks, weak or unreliable at others. The boundary between “useful” and “useless” is irregular, not linear.
- Fat layer of humans: Tom Blomfield predicts AI will automate many knowledge‑work tasks so that fewer humans are needed for operational execution; humans will remain as a thin oversight/judgment layer (and that layer could thin to near zero over long horizons).
- Productivity promise vs present confusion: Many firms see little short‑term gain and many projects fail, but startups deploying narrowly targeted AI solutions are scaling rapidly and showing big engagement and revenue gains.
- Organizational response matters: Ethan Mollick argues companies shouldn’t reflexively treat AI as a simple cost‑cutting tool. Instead, invest in organizational R&D, redesign workflows, and reallocate capacity to build new advantages.
- Practical roles for AI today: research assistant, first drafts for low‑identity tasks, editorial feedback, automation of repetitive spreadsheet/operational work, and rapid prototyping by small cross‑functional teams.
Notable quotes
- “The jagged frontier.” — Ludwig Siegeler, describing AI’s uneven strengths and weaknesses.
- “Right now, there’s a very fat layer of humans…and that layer gets thinner and thinner.” — Tom Blomfield on the future of human roles.
- “Forget bedside manner. Enter bedside scanner.” — Tom Blomfield, on AI in medicine.
- “To think of AI first as a human replacement, you’re wrong. But if you think it replaces your newcomers, you’re not just wrong, you’re out of your mind.” — teaser line summarizing a managerial view promoted in the series.
Concrete takeaways and recommended actions
- Experiment broadly and often: try AI on every part of your job for a short, intense test period to map the jagged frontier of what helps and what doesn’t.
- Keep human identity tasks for yourself: don’t outsource work that is central to your identity or where the thinking process is the primary value (e.g., first drafts for writers, deep creative thinking).
- Use AI as an editor/research assistant: write first drafts yourself, then use AI for summarizing literature, brainstorming, fact‑checking, and editing.
- Track hallucinations and trust thresholds: monitor where AI is confidently wrong and decide when the checking overhead outweighs the benefit.
- Reimagine work and org structure: invest in small R&D experiments (e.g., senior expert + engineer + domain person teams) to create new products and use AI capacity to expand capabilities rather than only cut headcount.
- Be transparent and reflective: record experiments, reflect on what the AI does well/wrong, and use that to redesign workflows.
Episode’s practical examples
- Voice‑clone demo: a party‑trick clone trained on Palmer’s writing and voice — quick to produce but raises ethical and identity questions.
- Blind writing test: Palmer vs AI (Claude) parody; close results in a blind taste test showed AI could produce coherent, conversational copy and sometimes fooled colleagues — revealing both the promise and personal unease.
- Organizational prototype: Mollick describes teams of three (senior subject‑matter expert, senior engineer, marketer/strategist) building new products in a week — illustrating rapid prototyping enabled by tooling and new workflows.
Why this episode matters
- It balances hype and skepticism: guests span Silicon Valley futurism (quick large gains and displacement) and newsroom/academic caution (jagged strengths, hallucinations, need for redesign).
- Actionable for managers and professionals: practical steps to experiment, protect core human work, and rethink how to capture AI gains without reflexive layoffs.
- Sets the season’s agenda: the episode frames the rest of the series as a journey toward practical deployments, organizational experiments, and lessons from places where AI is already changing jobs.
Quick checklist (for listeners who want to act now)
- Run a one‑week “use AI for everything” experiment at work to find concrete wins and failures.
- Separate tasks into: identity/core judgment (keep human), repetitive/operational (automate), and borderline (test with AI and measure checking cost).
- Start small R&D teams to prototype AI‑augmented workflows and products.
- Document errors/hallucinations and build simple verification workflows.
- Train colleagues on safe and effective prompt‑use and record outcomes.
End note: Palmer’s personal experiments show AI is already a competent research and editing assistant but can be unnerving when it imitates human voice or style convincingly. The pragmatic message: expect confusion, experiment deliberately, and redesign work to capture AI’s potential rather than just use it to cut costs.