Will work exist?
A reasoned guide to automation, gig labor, the four-day week, the universal-basic-income experiments, and the awkward question of meaning after the job.
Where work is, before AI gets to it
Three waves of automation, again
Every generation since the Luddites has watched a new technology threaten work. Power looms, internal combustion, electrification, the computer, the spreadsheet. Each wave displaced specific tasks; in aggregate, employment grew. The question for AI is whether the pattern holds.
Daron Acemoglu & Pascual Restrepo, in a series of papers (2018–24), distinguish automation (replacing labor) from augmentation (complementing labor). The mix matters: automation-heavy phases see wage stagnation and inequality; augmentation-heavy phases see broad-based gains. The 1985–2018 wave was, on their analysis, more automation than the rhetoric admitted.
The AI question is open. Generative AI looks more like augmentation in early studies (Brynjolfsson, Li, Raymond, 2023; Noy & Zhang, 2023; Peng et al. on Copilot, 2023). But the same tools, deployed end-to-end with agents, look much more like automation.
The honest answer for 2026 is: we don't know which wave AI will be. The institutional choices we make in the next ten years will shape it.
What the labor share has done
The remote experiment
The 2020 lockdowns ran a global natural experiment on remote work. Five years on, the new equilibrium is hybrid. Stanford's WFH Research (Bloom, Barrero, Davis) finds:
- ~28% of all paid working days remote in the US (2024)
- Hybrid-3 (3 office days/wk) the modal pattern in white-collar
- Productivity neutral to slightly positive on rigorous studies; preference asymmetric — workers value WFH at ~8% of wages
- Office occupancy stuck at ~50% of pre-2020 levels in major US cities
- Geography: secondary cities (Austin, Boise, Tampa) absorbed remote-first inflows
The downstream consequences — commercial real estate distress, downtown retail vacancy, school district financing — are still working through.
Which jobs, which tasks
The right unit of analysis is the task, not the job. Eloundou et al. (OpenAI, 2023) coded 950+ occupations against GPT-4 capabilities; ~80% of US workers had at least 10% of their tasks exposed; ~19% had at least 50% exposed. Highest exposure: tax preparers, writers, mathematicians, web designers.
| Occupation cluster | Task exposure | Likely shift |
|---|---|---|
| Tax prep · bookkeeping | Very high | Compression of mid-tier; higher-end advisory persists |
| Software development | High | Productivity multiplier, not displacement (so far) |
| Writing · editorial · marketing | High | Volume up, per-piece value down |
| Customer service | High | Tier-1 hollow-out; Tier-3 expert demand stable |
| Healthcare diagnosis | Augment-leaning | Imaging, triage, transcription · clinician judgement remains |
| Skilled trades · plumbing · electrical | Low | Wages rise; embodiment is the moat |
| Caregiving · early-childhood | Low | Demographic demand grows |
What platforms did to labor
Uber, DoorDash, Deliveroo, Upwork, Mechanical Turk turned millions of jobs into discrete tasks priced at the margin. The 2010s brought efficiency and flexibility; also algorithmic management, opacity, and worker classification fights. Prop 22 (CA, 2020), the EU Platform Work Directive (2024), the UK Supreme Court's Uber v Aslam (2021) all push back on bright-line independent-contractor status.
What we now know
- Platform pay is highly variable; mean often above legal wage, median below
- Most gig workers also have a primary job
- Algorithmic management produces fatigue and arbitrary discipline
- Worker organizing is harder, not impossible (App Drivers & Couriers Union)
| Daron Acemoglu | MIT · Nobel 2024 · institutions, automation, Power and Progress (2023) |
| David Autor | MIT · the canonical "polarization" thesis; recent work on AI & middle skills |
| Daniel Susskind | Oxford · A World Without Work (2020) |
| Erik Brynjolfsson | Stanford Digital Economy Lab · the Turing Trap; productivity J-curve |
| Carl Benedikt Frey | Oxford Martin · 2013 "47% at risk" paper; corrections since |
| Mary Gray & Siddharth Suri | Ghost Work (2019) · the human labor inside automation |
| Anna Salomons | Utrecht · long-run wage compression evidence |
| Aaron Benanav | Automation and the Future of Work (2020) · skeptic case |
What the trials say
Universal Basic Income, once a fringe idea, has been tested at modest scale in Finland (2017–18, ~2,000 participants), Stockton CA (2019–21, $500/mo), GiveDirectly Kenya (2017–, full-saturation), Kela's earlier negative-income-tax pilots, OpenResearch (Sam Altman's, results 2024). Results are interesting and limited.
| Study | Design | Headline finding |
|---|---|---|
| Finland 2017–18 | €560/mo, unemployed | No employment effect; well-being up |
| Stockton SEED 2019–21 | $500/mo, 125 ppl | Full-time employment ↑; volatility ↓ |
| GiveDirectly Kenya 2017– | $22/mo, ~10yr saturation village | Asset accumulation, business formation |
| OpenResearch 2024 | $1k/mo, 3yr, 1k US adults | Modest reduction in work hours; spending on basics |
forecast By 2035 expect 5–10 OECD countries to run formal pilots; full UBI implementation remains politically distant.
Working less, on purpose
4 Day Week Global's UK pilot (2022, 61 firms, ~2,900 employees) reported flat-or-improved revenue, lower burnout, lower turnover. 92% of firms continued. Iceland's 2015–19 trial (1.3% of workforce) had similar results. Spain, Portugal, Belgium have national-scale pilots underway.
The mechanism: a 100-80-100 commitment — 100% pay, 80% time, 100% output — forces meeting reform, deep-work protection, and elimination of low-value work. Critics note selection bias (high-knowledge-work firms self-select).
If work shrinks, what fills the room?
Susskind's World Without Work ends not with optimism or pessimism but with a problem: leisure is hard. Aristotle thought leisure was where philosophy lived. Modern industrial workers, mostly, have not used leisure that way.
The honest sociology says: identity, status, social ties, and time-structure are the four things work currently provides. Replacing it requires explicit institutions — community, creative practice, learning, civic life. Bregman, Standing, Frase argue this is the actually-hard problem, distinct from the income problem.
If we automate the wages, we still have to build the meaning somewhere else.
Recommended source
Lex Fridman × Daron Acemoglu
Three-hour conversation on Power and Progress, automation, AI, and institutions. The clearest articulation of the augmentation-vs-automation distinction available.
youtube.com/@lexfridman →PBS Idea Channel · Future of Work series
Shorter, accessible takes; pair with Andrew McAfee's MIT lectures on YouTube.
youtube.com/@PBSIdeaChannel →Three plausible 2050s
Indicators · 2026–2030
- Labor-share trend in OECD (BLS, ONS, Eurostat) — falling further?
- Productivity J-curve · does measured TFP finally rise?
- Worker classification rulings (US PRO Act, EU Platform Directive)
- Frontier-AI agent-capability benchmarks (METR, GAIA) saturating?
- UBI / NIT-style policies becoming live ballot questions
- Office occupancy data (Kastle, Placer.ai) — recovery or new normal?