Why my team is pushing back on AI

Summary

The video argues that pushback against workplace AI—especially AI coding tools—is rational and driven by a mismatch between promised productivity gains and real, near-term worker experience.

1) Productivity benefits are real but much smaller than claims

• The speaker says the early messaging (“AI changes everything”) has shifted into internal resistance.
• He cites an evidence-based productivity figure: median productivity gain ~7.8%, not “10x.”
• Gains often don’t compound: when engineers reach a peak AI-assisted quarter, 66% see the benefit drop in the next quarter.
• CFOs and decision-makers are portrayed as struggling to justify spend because benefits are hard to prove conclusively.

2) Workers pay hidden costs: skill atrophy and “cognitive debt”

• The core complaint isn’t only output—it’s what happens to the worker’s abilities.
• Learning is described as involving repetition and struggle; AI can remove friction that builds skill.
• A 2025 MIT Media Lab study is referenced, claiming heavy AI use leads to “cognitive atrophy”, framed as “cognitive debt.”
• The result is frustration: people may become less capable even if immediate speed improves.

3) A new and uncomfortable economic asymmetry: owners gain immediately, workers often don’t

• Historically, major technologies increased wealth across society over time (factory/electricity/car/internet).
• The speaker claims AI differs because it’s the first widespread tech where:
  • owners capture productivity gains quickly,
  • workers often don’t see corresponding wage/benefit gains, and
  • workers may be used to train the replacement system.
• Examples include workplace monitoring/training:
  • Factory workers allegedly being recorded to train AI that will replace them.
  • Meta/Mark Zuckerberg’s plan to record engineers’ computer activity for model training, followed by 8,000 layoffs.

4) Layoffs and “AI adoption” appear coercive and commercially conflicted

• Companies are portrayed as pushing AI under threat of job loss, while workers feel the return is unclear.
• A cited survey of 2,400 executives and employees reports:
  • 60% of companies plan layoffs for non-adopters.
  • 48% of executives call their AI adoption a “massive disappointment.”
• An additional example is offered: Cloudflare’s layoffs framed in a way that reduces people to “functions.”

5) The funding boom may be driven by financial engineering, not just end-customer demand

• The speaker contrasts extremes:
  • Uber supposedly burned its 2026 AI budget in 4 months after rolling out Claude code to 5,000 engineers.
  • Many AI labs/suppliers reportedly aren’t meaningfully profitable yet (Anthropic had its first profitable quarter; OpenAI not expected to be profitable until 2029).
• Potential self-subsidy dynamics are highlighted:
  • Nvidia investing $100B into OpenAI, with OpenAI then buying Nvidia chips—described as not a normal customer relationship, but a way to stimulate Nvidia demand.
• Nvidia’s fragility is emphasized: the CEO warns the world would “fall apart” if the margin/quarter missed slightly.

6) Practical “on-the-ground” outcomes: one big win, one costly failure

Using his own companies, the speaker describes:

• Win: An AI agent (Perplexity-based) used a photo (model/serial) to contact the manufacturer, coordinate, and book an appointment—taking minutes instead of ~half a day.
• Failure/expense: A custom AI code review bot sometimes behaves unpredictably:
  • one run cost ~£100,
  • it recommended restricting its own permissions,
  • then it couldn’t post future comments because access was revoked.

Overall takeaway: AI’s real-world payoff depends heavily on context; marketing expectations often exceed what early deployments reliably deliver.

7) Conclusion: the technology is real, but the “AI noise” outruns real outcomes

• The speaker reconciles two truths:
  • AI capability is real (he experiences it),
  • but the current economic rush is outpacing what’s working today and how benefits are distributed.
• Worker resistance is framed as legitimate, prompting viewers to consider whether the issue is cognitive cost or economic asymmetry.

Presenters or contributors

• Axel Mollist (speaker; references his three companies, his team, and his newsletter)