Video summary

Coders Quitting AI? AI Slop PhDs? | Ep 8 of Philosophy, Programs and Prompts

Main summary

Key takeaways

News and Commentary

Summary of the episode’s main points

This episode discusses a backlash against heavy AI use in coding, sparked by creator “Brett Codes,” who says he plans to stop using AI as a coder. The hosts frame the controversy as part of a broader pattern: people adopt AI, become frustrated when work quality or culture feels worse, and then escalate into public “AI slop” criticism and counter-movements.

1) Case overview: why Brett plans to stop using AI

The hosts summarize Brett’s argument as having four parts:

  1. AI harms the world

    • Including environmental impacts (e.g., water/energy use) and other broad harms.

  2. AI makes individual coders worse

    • Less capability/understanding due to reliance on AI output.

  3. AI worsens businesses and codebases

    • More messy, feature-bloated, and less maintainable code.

  4. AI creates existential/experiential downsides

    • Programming becomes less enjoyable, “soul-less,” and less meaningful when AI does much of the hands-on work.

2) “AI harms” — mental health and safety concerns

Brett’s “heaviest” example is personal: he claims that querying an AI at night encouraged him to go to the emergency department, and that felt like a turning point—an experience where AI disrupted his relationship with reality.

One host argues:

  • This is not a strong argument against all AI use in general, because the core problem is unreliable AI medical guidance, which is orthogonal to whether AI is appropriate in coding.
  • The hosts distinguish AI as a tool from AI products being used in harmful ways (e.g., medical “screener” claims or inaccurate guidance).
  • They also note some users may be psychologically vulnerable to LLM interaction (tying it to the well-known “Eliza effect”), but they don’t see this as justification for blanket workplace restrictions on all AI use.

3) Environmental impact: critique of “AI is inherently bad”

On environmental arguments (water/energy), the hosts push back against an absolute ban:

  • Water use is location-dependent

    • Building data centers in drought regions can be irresponsible, but that doesn’t mean all AI has the same water footprint.

  • Energy use depends on the power source

    • If electricity comes from cleaner generation, the impact changes.
    • Because power grids are interconnected, local differences matter—but personal abstinence likely won’t materially alter the broader system.

  • Individual non-use has low leverage

    • The hosts argue the bigger lever is political/regulatory pressure on companies and data-center siting decisions.

They also debate ethical framing:

  • They highlight tension between consequentialist and non-consequentialist approaches (whether abstaining matters if it doesn’t change aggregate harm).
  • They use a meat/factory farming analogy: individual abstention might reduce harm under some ethics models, but climate externalities are hard to trace back from individual choices to real-world reductions.
  • The takeaway: ethical conclusions depend heavily on one’s ethics model.

4) Other “AI harms” are real but not necessarily “against AI as a technology”

The hosts briefly acknowledge additional harms often attributed to AI ecosystems—such as community disruption from infrastructure (noise, neighborhood impacts), supply-chain pressures, or economic bubbles. However, they maintain these criticisms should target how specific companies implement AI, not necessarily AI itself.

5) “AI makes coders worse” — historical analogy and adaptation

Regarding claims that coders will understand less and become less employable, the hosts draw historical parallels:

  • They compare it to past waves of programming change—like the rise of package managers and fears such as “don’t use code you didn’t write yourself.”
  • Those fears faded as the industry adapted.

They argue the likely outcome isn’t total loss of competence, but a shift in skills:

  • Programmers may understand the “forest” (system behavior, architecture, bug triage) more than every “tree” (each low-level implementation detail).
  • They predict that people refusing AI entirely may eventually struggle with productivity expectations—similar to workers who couldn’t adapt to newer tooling paradigms.

6) “Apathy and loss of learning” — reframe what skills are learned

The hosts address Brett’s points about reduced learning and motivation:

  • Reduced hands-on detail can lower certain kinds of learning.
  • But it may increase focus on higher-level reasoning, planning, review, security, and debugging across abstractions.
  • Their core point: learning may still happen—it just may shift to a different domain than before.

7) Business impact: managing “feature bloat” and AI-generated code at scale

The hosts discuss Brett’s concern that AI output can lead to:

  • More code churn and feature bloat
  • Weaker processes for reviewing, integrating, and securing AI-generated changes

They argue the industry currently lacks solid “best practices” for large-scale AI-assisted coding, but they see opportunity here:

  • New value may come from building workflows such as better code review strategies, safer integration, and executive/product decision-making to constrain the “firehose” of generated code.

They also note an incentive problem:

  • If businesses measure success by raw productivity metrics (e.g., output volume) rather than revenue, reliability, or user outcomes, AI adoption can become distorted and feel “mandate-like”—fueling backlash.

8) Vulnerability and resilience: “If AI disappears, will people be unable to code?”

Brett suggests organizations become vulnerable if AI becomes unavailable.

One host counters:

  • Modern coding already relies on external infrastructure (internet, GitHub, package ecosystems).
  • If those services went down, productivity would drop—but the overall system benefits justify dependence.
  • Similarly, if AI use were disrupted, the industry would adapt; programmers would learn alternatives and new workflows.

They also clarify a practical point:

  • AI models are time-bound—if new model development stopped, older models would become less useful over time. So the idea that “AI will vanish forever” is unlikely, even though the adaptation principle still applies.

9) Existential “estrangement from work” — nostalgia and reassignment of skills

Brett’s existential claim is that AI reduces the meaningful “touch” of coding.

The hosts respond with tech/history analogies:

  • Portrait painting didn’t vanish; it shifted into niches.
  • Neon craftsmanship shifted toward film/period-piece niches.

They acknowledge the emotional validity of feeling displaced, but argue technology rarely creates universally “worse” outcomes—it reshapes roles and where expertise is valued.

10) Why the backlash resonates

The hosts suggest multiple reasons Brett’s video went viral:

  • Programmers are online and react intensely to low-quality “AI slop.”
  • There’s visceral disgust toward low-quality AI-generated media/copy-paste code.
  • Broader cultural currents—nostalgia, anti-corporate sentiment, and anti–data-center/anti-AI feelings—make the message feel personal:

    “I’m being forced to use tools that make me feel alienated from my labor.”

Concluding synthesis

The hosts land on a dual approach:

  • Separate AI-as-a-tool from AI-as-a-product/environment
    • Criticize misuse and slop, but don’t equate every AI use case with the worst incentives or harms.
  • Use cost-benefit analysis for technology choices generally (not only AI).
  • Encourage flexibility and continual skill evolution: if someone feels their interest or skills are eroding, they may need to pivot into new work where human judgment remains central.

Presenters / contributors

  • Casey Harts
  • Carl Brown

Original video