read
articles
VIP VIP Avital Balwit

VIP VIP Avital Balwit

📝 ARTICLE INFORMATION

  • Title: VIP Avital Balwit
  • Person: Avital Balwit
  • Position: Chief of Staff to CEO at Anthropic
  • URL: avitalbalwit.com
  • Date: October 15, 2025
  • Word Count: Approximately 6,500 words

🎯 HOOK

At 25, the Chief of Staff to Anthropic’s CEO confronts an uncomfortable truth: the AI she helps build will likely make her job (and yours) obsolete within five years, and she’s not sure that’s actually bad.

💡 ONE-SENTENCE TAKEAWAY

Avital Balwit argues that AGI-induced unemployment might enable human flourishing if we eliminate the shame of joblessness, provide material security, and learn to find meaning in activities we do badly, preparing for a future where excellence belongs to machines.

📖 SUMMARY

Avital Balwit writes from the strange position of watching her profession disappear while helping to build its replacement. As Chief of Staff to Anthropic CEO Dario Amodei, she sees each model iteration grow more capable, more general, more able to do what she once took pride in doing: writing 2,000 words per hour, synthesizing information, generating cogent content across topics. She calls this skill “cutting blocks of ice from a frozen pond” arguably obsolete.

Her flagship essay “My Last Five Years of Work” examines whether humans can be happy without employment. The question isn’t hypothetical. She expects AI to excel at any online work within five years: copywriting, tax preparation, customer service, software development, contract law. Anything involving reading, analyzing, synthesizing, and generating content faces automation. Physical work lags behind: electricians, gardeners, plumbers, jewelry makers, hair stylists may work decades longer. Regulated industries like medicine will retain humans longer. “Nostalgic jobs” where human relationship matters (counselors, preschool teachers, priests) will persist because consumers value the human doing the work.

The conventional wisdom says unemployment destroys wellbeing. Balwit interrogates this assumption through research synthesis. Studies show unemployed people report worse mental and physical health, but causality confounds interpretation: sick people lose jobs more easily. When researchers isolate unemployment from financial stress and personal shame (pandemic layoffs, plant closures) the picture shifts. Canadian workers temporarily laid off during COVID reported their experience as “forced vacation,” appreciating the break from work stress. Depression from plant closures was lower than from individual layoffs, suggesting shame matters more than joblessness itself.

Historical perspective challenges our assumptions about necessary work hours. Work weeks have declined over 150 years as countries got richer. The 8-hour day and weekend were labor movement triumphs for human wellbeing. Why assume 40 hours weekly is optimal? Why believe less work was better in the past but would be worse now?

The retirement literature offers contradictory findings. Some studies show retirees report better mental and physical health. Others show the opposite: increased mobility difficulties, illness, mental health decline. One robust finding: happiness follows a U-shape by age, with people 60-75 reporting highest satisfaction; precisely when they’re typically retired. Women entering the formal labor force from 1890 (18% participation) to 2016 (57%) didn’t report increased happiness, suggesting formal employment isn’t the happiness key. Aristocrats historically filled unemployment with social rituals, hobbies, studying, art, writing…and seemed hedonically fine.

Balwit’s synthesis suggests three factors determine whether unemployment makes people miserable: financial security (addressable through universal basic income), shame (eliminated when everyone’s unemployed), and time use (research shows moderate discretionary time is optimal, but “social” and “productive” leisure can extend indefinitely without harming wellbeing).

She frames the future through two lenses. First, skill displacement isn’t about matching the best human. It’s about replacing the human who would otherwise do that task. Second, we must practice doing things we’re bad at for joy rather than excellence. An AI researcher she knows prepares for post-AGI by taking up jiu-jitsu and surfing, “savoring the doing even without excellence.” This is preparation: learning to fill days from joy rather than need, even when machines do everything better.

The essay closes with an optimistic twist: if we build aligned AGI capable of replacing human work, it will also be capable of helping us solve the problems it creates. If solutions to unhappiness and purpose-loss exist and can be found through intelligence, these superhuman systems should help us find them.

Beyond the flagship essay, Balwit’s background spans creative writing (short stories in The Massachusetts Review, Prairie Fire; poetry in The Atlantic’s contest), political theory, and cognitive science. She campaign-managed a House race, did grantmaking in AI safety and biosecurity, and researched at Oxford before joining Anthropic. Her writing bridges technical AI understanding with humanistic concern, embodying the tension of building the technology that might obsolete her.

🔍 INSIGHTS

Core Insights

  • Replacement Threshold Matters More Than Peak Performance: Economic displacement happens when AI surpasses the human who would do that task, not when it matches the best human. Most workers aren’t writing award-winning books or patenting inventions—they’re doing competent information synthesis.
  • Shame Confounds Unemployment Studies: Research showing unemployment harms wellbeing conflates financial stress, personal shame, and joblessness itself. Pandemic layoffs and plant closures (“blameless” unemployment) show markedly different psychological outcomes.
  • Work Hours Are Historically Contingent: The 40-hour week isn’t natural law. Work hours declined over 150 years as wealth increased. Weekends and 8-hour days were fought for as wellbeing improvements. Current standards aren’t optimal endpoints.
  • Discretionary Time Has Optimal Structure, Not Optimal Amount: Research suggests moderate discretionary time maximizes wellbeing, but distinguishes between “social/productive” leisure (which can extend indefinitely) and “solo/unproductive” leisure (which can become excessive). How you spend time matters more than quantity.
  • Excellence Isn’t Required for Meaningful Activity: Ballet dancers who’ll never be prima ballerinas still find joy in movement. Friends counsel each other despite professional therapists being more skilled. Doing things badly for joy or relationship value is practice for post-AGI living.
  • The Model Release Cycle Masks Continuous Progress: Language models improve discontinuously…appearing to plateau while new versions “bake” for months. What looks like stagnation actually means the next jump is in the oven.
  • Nostalgic Jobs Have Human-Relationship Premium: Some work persists not because humans do it better but because consumers value human involvement: counselors, doulas, childcare, teachers, priests. The relationship IS the value, not just the outcome.

How This Connects to Broader Trends/Topics

  • AI Safety Movement: Inside perspective on how frontier lab employees think about transformative AI impacts; bridges technical development with social consequences.
  • Universal Basic Income Debates: Provides framework for evaluating UBI success criteria beyond financial security. It must also address shame and time-use patterns.
  • Future of Work: Challenges both techno-optimist (AI creates new jobs) and techno-pessimist (unemployment is inherently bad) positions with nuanced synthesis.
  • Effective Altruism: Background in EA grantmaking and FHI research informs approach to long-term impacts and evidence-based policy.
  • Work Identity Crisis: Examines Protestant work ethic’s grip on meaning-making as AGI threatens to eliminate the employment-identity link.

🛠️ FRAMEWORKS & MODELS

Three-Factor Unemployment Wellbeing Model

  • Explanation: Whether unemployment harms wellbeing depends on three separable factors: financial security, social shame, and time use structure.
  • Components:
    • Financial Security: Material needs must be met (addressable through UBI or transfers)
    • Shame Elimination: Unemployment must not signal personal failure or set one apart from peers (achieved through universal automation)
    • Time Use: Discretionary time must be spent on social/productive activities rather than solo/unproductive ones
  • Application: Policy design should target all three factors simultaneously. UBI alone insufficient without addressing shame and providing time-use structures.
  • Significance: Reframes automation anxiety from “will people have jobs?” to “can we create conditions where joblessness doesn’t harm wellbeing?”
  • Evidence: Pandemic layoffs showed lower distress than normal unemployment; plant closures showed less depression than individual layoffs; retirement literature shows mixed results correlating with these factors.

Replacement Threshold vs. Peak Performance

  • Explanation: Economic displacement occurs when AI can outperform the marginal worker, not the exceptional worker.
  • Application: Stop asking “can AI write award-winning novels?” and start asking “can AI do competent content synthesis better than the person currently doing it?”
  • Significance: Shifts timeline dramatically as many jobs face near-term displacement even though AI hasn’t matched human peak performance.
  • Critical Insight: Knowledge workers engage in denial by focusing on ever-shrinking areas where AI struggles rather than ever-expanding areas where it matches or exceeds human performance.

Nostalgic Jobs Theory

  • Explanation: Certain occupations will persist not because humans perform better but because consumers value human involvement inherently.
  • Categories:
    • Relationship-based care (counselors, doulas, elderly caretakers)
    • Child-rearing (babysitters, preschool teachers)
    • Spiritual guidance (priests, religious leaders)
    • Intimate services (sex workers, therapists)
  • Application: Career planning should consider whether job value comes from output quality or human relationship itself.
  • Limitations: Real wages in nostalgic jobs might not sustain full labor force participation; unclear if relationship-premium persists when AI becomes very capable.

Joy-Based vs. Excellence-Based Activity Framework

  • Explanation: Post-AGI life requires shifting from doing things because you’re good at them to doing things for hedonic, relational, or virtue reasons despite being outperformed.
  • Three Justifications:
    • Hedonic: Ballet brings joy even if you’ll never be prima ballerina
    • Relational: Counseling friends matters even if professional therapists are more skilled
    • Virtue/Moral: Writing betters you even if others write better
  • Application: Practice now by taking up activities you’re notably worse at than others (jiu-jitsu, surfing) and savoring the doing without excellence.
  • Significance: Psychological preparation for world where machines outperform humans at everything economically valuable; meaning must come from process, not outcome.

Discretionary Time U-Curve

  • Explanation: Too little discretionary time harms wellbeing (obvious), but too much can also harm wellbeing…unless spent on social/productive activities.
  • Application: Post-automation society must create structures for social and productive leisure, not just eliminate work.
  • Significance: Challenges both “leisure is unambiguously good” and “idleness breeds misery” positions; structure and content of free time matter more than quantity.

💬 QUOTES

  • “I am 25. These next five years might be the last few years that I work.”

    • Context: Opening of “My Last Five Years of Work,” establishing the essay’s provocative premise from someone building the technology causing her obsolescence.
    • Significance: Encapsulates the insider’s dilemma, working to build the system that will eliminate the need for your work.

  • “The economically and politically relevant comparison on most tasks is not whether the language model is better than the best human, it is whether they are better than the human who would otherwise do that task.”

    • Context: Explanation of replacement threshold versus peak performance framework.
    • Significance: Shifts entire displacement debate from exceptional workers to marginal workers, dramatically accelerating expected timelines.

  • “Freelance writing was always an oversubscribed skillset, and the introduction of language models has further intensified competition…a skill which, like cutting blocks of ice from a frozen pond, is arguably obsolete.”

    • Context: Personal reflection on her once-valuable ability to write 2,000 words per hour.
    • Significance: Vivid metaphor for how entire skillsets can become obsolete overnight…ice cutting was once essential, now it’s a curiosity.

  • “The shared goal of the field of artificial intelligence is to create a system that can do anything. I expect us to soon reach it.”

    • Context: Straightforward statement of AGI timeline expectations from someone with direct observation of frontier development.
    • Significance: Unhedged confidence from insider perspective; not speculating but reporting what she sees in model iterations.

  • “A renowned AI researcher once told me that he is practicing for post-AGI by taking up activities that he is not particularly good at: jiu-jitsu, surfing, and so on, and savoring the doing even without excellence.”

    • Context: Near the essay’s close, offering concrete example of psychological preparation.
    • Significance: Demonstrates that even those building AGI are thinking seriously about adapting to a world where human excellence is obsolete.

  • “If we believe there are solutions to unhappiness or a feeling of a loss of purpose, and that these solutions can be found with intelligence, then we should expect these systems to be able to help us find them.”

    • Context: Closing argument about aligned AGI helping solve the problems it creates.
    • Significance: Optimistic turn that reframes AGI from pure threat to potential collaborator in human flourishing.

⚡ APPLICATIONS

Career Strategy in the AI Age

  • Timing Assessment: If you work primarily online doing information synthesis (writing, analysis, research, coding), expect 3-7 year horizon for significant AI competition.
  • Refuge Identification: Move toward physical work requiring varied delicate movements (electrician, plumber, jewelry maker) or nostalgic jobs where human relationship is the product (therapist, teacher, doula).
  • Skill Futility Recognition: Stop optimizing for skills AI will soon exceed; start identifying activities you value independent of comparative performance.

Psychological Preparation

  • Excellence Detachment Practice: Take up activities you’re bad at (martial arts, artistic pursuits, new sports) and practice enjoying the process without comparing to others.
  • Meaning Reorientation: Identify what you do for hedonic (joy), relational (connection), or virtue (self-improvement) reasons rather than economic or status reasons.
  • Social Time Structuring: Since research shows social/productive leisure sustains wellbeing indefinitely, build habits around community activities, creative projects, physical movement with others.

Policy Development

  • Triple-Threat Design: UBI or transfer programs must address financial security, shame elimination, and time-use structure simultaneously because single-factor solutions are insufficient.
  • Shame Normalization: Public messaging should emphasize that automation-induced unemployment is blameless and universal, fundamentally different from individual job loss.
  • Leisure Infrastructure: Invest in community spaces, makerspaces, sports facilities, arts programs that enable social/productive discretionary time use.
  • Nostalgic Job Subsidy: Consider whether to subsidize relationship-based occupations (teaching, caregiving, counseling) to maintain human employment in high-value social roles.

Individual Preparation

  • Financial Planning: Assume 5-10 year horizon for major employment disruption; build savings or recession-proof income streams.
  • Community Building: Strengthen local ties, join activity groups, develop social infrastructure that persists independent of employment.
  • Identity Diversification: Reduce attachment to professional identity; cultivate self-concept based on relationships, hobbies, values rather than job title.
  • Proactive Experimentation: Don’t wait for automation to force change. Instead, experiment now with reduced work hours, sabbaticals, extended time off to test how you handle unstructured time.

Organizational Response

  • Remote Work Automation: Recognize that remote-first organizations face earliest disruption since online work is most automatable; plan for AI augmentation or replacement.
  • Human-Touch Premium: For customer-facing roles, evaluate whether customers value human interaction itself or just the outcome; invest in human relationships where they matter.
  • Transition Support: Design programs helping employees develop non-work identity and time-use skills alongside traditional severance.

📚 REFERENCES

Academic Research Cited

  • Sullivan and von Wachter (2009); Eliason and Storrie (2009); Browning and Heinesen (2012): Large mortality effects after job displacement (50-100% increases)
  • Rege et al. (2009): Smaller mortality effects (10-15% increases)
  • Kuhn et al. (2009); Black et al. (2015); Salm (2009): Negligible or zero health effects from unemployment
  • Spanish Construction Industry Study: 15% increase in poor health reports, 33% increase in mental disorder reports from unemployment
  • Schieman, Mai, and Kang (2020): “A Forced Vacation? The Stress of Being Temporarily Laid Off During a Pandemic” found lower distress among temporarily laid-off workers
  • Brand, Levy, and Gallo: “Effects of Layoffs and Plant Closings on Depression Among Older Workers” found gender differences in depression from layoffs vs. closures
  • Sharif, Mogilner, and Hershfield: “Having Too Little or Too Much Time Is Linked to Lower Subjective Well-Being” U-shaped relationship between discretionary time and wellbeing
  • Dave, Rashad, and Spasojevic: “The Effects of Retirement on Physical and Mental Health Outcomes” negative retirement effects
  • French GAZEL Study: Retirement reduces fatigue and depression, especially among those with chronic diseases
  • Stevenson and Wolfers: Female happiness declined 1970s-2000s despite increased labor force participation

Historical & Economic Context

  • Our World in Data: Working hours decline over 150 years; female labor force participation trends
  • Scaling Laws Research (Kaplan et al., 2020): “Scaling Laws for Neural Language Models”—predictable improvement with more compute, data, parameters
  • Korinek (2023); Susskind (2017): AGI economic models and technological unemployment scenarios

Cultural References

  • Iain Banks’ Culture Series: Post-scarcity civilization where “the urge not to feel useless” drives Contact Section’s galactic interventionism
  • Aristocratic Unemployment: Historical comparison to landed gentry filling time with social rituals, hobbies, intellectual pursuits

Professional Background

  • Future of Humanity Institute (Oxford): Research on transformative AI
  • Rhodes Scholarship: Background in political theory and cognitive science
  • Anthropic Position: Chief of Staff to CEO Dario Amodei; direct observation of frontier model development
  • Tyler Cowen Collaboration: Co-authored essay in The Free Press on AI’s impact on human meaning

Creative Writing Portfolio

  • Short Stories: Published in The Massachusetts Review, Prairie Fire, Pop Up UK, Meetinghouse Magazine, Chillfiltr Review
  • Poetry: The Atlantic’s 2020 Instagram Poetry Contest winner; published in Coastal Shelf
  • Children’s Literature: Book on octopus cognition (Pop Up UK, 2021)

⚠️ QUALITY & TRUSTWORTHINESS NOTES

Strengths

  • Insider Perspective: Direct daily exposure to frontier AI development provides empirical grounding for timeline predictions.
  • Research Synthesis: Cites peer-reviewed studies across psychology, economics, public health; engages with contradictory evidence.
  • Epistemic Humility: Acknowledges uncertainty (“If I’m right…”), examines counterarguments, notes when research shows mixed results.
  • Personal Disclosure: Explicitly states writing in personal capacity, not representing Anthropic; transparent about position influencing perspective.
  • Interdisciplinary Approach: Combines technical AI understanding, psychology research, economic analysis, historical comparison, and personal reflection.

Considerations

  • Timeline Confidence: “I expect us to soon reach it” regarding AGI is unhedged prediction; reasonable people disagree significantly on timelines.
  • Selection Bias in Studies: Retirement and unemployment research shows contradictory results; synthesis may overweight studies supporting her thesis.
  • UBI Assumption: Analysis assumes material needs can be met through transfers; this is contested policy assumption, not established fact.
  • Alignment Assumption: Closing optimism about aligned AGI solving problems it creates depends heavily on successful alignment…far from guaranteed.
  • Generalization from Privileged Position: As Rhodes Scholar at elite AI company, her financial security and social capital may not reflect broader population experience with job displacement.
  • Aristocrat Comparison Limitations: Historical aristocrats lived in societies with extreme inequality, different social structures; analogy may not transfer.

Accuracy Assessment

  • AI Capabilities: Characterization of current LLM abilities (competent content generation, passable text analysis) aligns with public Claude/GPT-4 performance.
  • Scaling Laws: Description of predictable improvement from compute/data/algorithms matches mainstream ML understanding.
  • Model Development Process: “Baking a cake” analogy (pretraining, post-training) accurately describes current LLM production pipeline.
  • Labor Market Vulnerability: Remote work / knowledge work facing earliest automation is consensus view among economists studying AI impact.

Trust Indicators

  • External Validation: Featured by Tyler Cowen (co-authored follow-up), discussed widely in tech/EA communities, media coverage in Fortune/Yahoo.
  • Academic Grounding: Rhodes Scholar with FHI research background; understands standards of evidence.
  • Intellectual Honesty: Examines studies contradicting her thesis; notes limitations of comparisons; acknowledges “this may sound self-serving” when discussing AGI helping solve problems.

Find this work at: avitalbalwit.com | @AvitalBalwit on X | Chief of Staff to CEO at Anthropic

Crepi il lupo! 🐺