Wes Roth critiques Cal Newport’s recent comments on AI progress, arguing that Newport’s claims about a slowdown in AI advancements are factually incorrect and overlook major breakthroughs in coding agents and new AI models. Roth provides evidence of rapid AI development, real-world impact, and economic growth, challenging Newport’s skepticism and inviting further discussion.
The video is a critical response by Wes Roth to Cal Newport’s recent commentary on the state of AI progress, particularly Newport’s reaction to Matt Schumer’s viral article, “Something Big Is Happening.” Roth expresses surprise and confusion at Newport’s takes, arguing that Newport’s depiction of AI progress is not just a difference of opinion but factually incorrect. Newport claims that AI progress slowed after 2025, especially after the shift from pre-training scaling to post-training and inference-time compute, whereas Roth insists that the most dramatic leaps in AI capability actually occurred during and after this period, especially with the introduction of “thinking” models and coding agents.
Roth challenges Newport’s assertion that the most significant progress happened between GPT-2, GPT-3, and GPT-4, and that things slowed down after that. He presents benchmarks and real-world examples to show that the real exponential improvements came with the newer models, such as Claude Opus 4.6 and others, which rapidly saturated benchmarks like ARC-EGI and began solving previously unsolved mathematical problems. Roth also points out that Newport’s dismissal of the impact of coding agents is misguided, as these tools have enabled users—including himself—to build complex applications with minimal human intervention, something Newport claims is not happening in practice.
A major point of contention is Newport’s reliance on feedback from professional programmers to gauge the impact of AI coding agents. Roth argues that this is a classic mistake, referencing “The Innovator’s Dilemma,” and points out that disruptive technologies are often underestimated by experts who compare them to the best existing alternatives, rather than appreciating their accessibility and utility for non-experts or new users. Roth provides personal anecdotes and broader industry examples to demonstrate that AI coding agents are already transforming workflows and enabling new kinds of productivity.
Roth also disputes Newport’s claim that making AI good at coding was not a deliberate strategy to accelerate AI development. He cites public statements and research from Google DeepMind, Anthropic, and Sakana AI, showing that AI agents are indeed being used to optimize and even help design the next generation of AI models and infrastructure. He references Alpha Evolve, which improved Google’s data centers and chip design, and notes that AI-driven research has led to breakthroughs in mathematics, materials science, and other fields—contradicting Newport’s assertion that AI agents cannot meaningfully contribute to AI research or recursive self-improvement.
Finally, Roth addresses Newport’s skepticism about the economic impact and market adoption of AI, pointing to explosive growth in AI company valuations, massive infrastructure investments by tech giants, and rapidly increasing revenue for leading AI labs. He argues that AI is not just finding “niche” applications but is already transforming major sectors like software, customer service, and content creation. Roth concludes by expressing genuine confusion at Newport’s perspective, inviting viewers to weigh in and clarify if he is missing something, and reiterates his respect for Newport while firmly disagreeing with his analysis.