Matt Shumer on his 50M-view AI essay: 'I originally wrote this for my parents'

Summary

Matt Shumer's essay on AI's trajectory has reached 50 million views. He wrote it originally to explain the moment to his parents during Super Bowl weekend after struggling to find anything that was both technically accurate and accessible to non-technical audiences. Dario Amodei's essays are rigorous but require tech fluency. Shumer aimed for something simpler: signal that something significant is happening and ask people to think about what it means for their own industry.

Shumer is careful to hedge his central claim. He doesn't argue that AI will transform every field overnight, and he acknowledges the gap between capability and real-world deployment. His father is a lawyer. AI can draft and review documents, functioning as a junior associate, but won't argue cases in court for years. In legal, partners may thrive while junior lawyers starting today face disruption. The insight is contextual. Pay attention, because the timeline matters for your specific role.

Shumer entered AI in 2019 after dropping out to build a VR startup, realizing he'd regret not pivoting immediately. The progression he's witnessed is stark. Models in 2019 couldn't write a coherent sentence. Now, he's had early access to GPT-5.3 Codex. He handed it a full spec and asked it to deploy autonomously without checking back. It did. That level of capability changes how he thinks about hiring, product strategy, and where value will accrete.

On the writing process itself, Shumer used Claude iteratively, not Codex, which he views as superior for engineering but worse for prose. He dumped years of AI research essays into the model, overlaid his own positions on what he agrees and disagrees with, spoke his thoughts aloud for an hour, and iterated for hours. Only then did he write a first draft himself, bringing the model back for critique and wording. The collaboration is back-and-forth refinement, not a one-shot prompt. Critics who dismiss it as "AI-written" miss the point. The viral reach proves the tool works as a multiplier on human judgment.

For new graduates facing an uncertain job market, Shumer offers no clean answer. His advice: learn the tools, because they are the baseline now. Choose industries that will take longer to disrupt so you can entrench yourself first. Be adaptable. The principle is clear: treat AI proficiency the way you'd treat a calculator, except amplified a thousand times over.

On his own portfolio, Shumer is backing infrastructure and interoperability plays. He's invested in chip companies like Etched and recently in what he calls "the rails"—the connective tissue that lets models communicate and break out of their boxes. Recent investments include AgentRelay and AgentMail, companies he views as critical for proliferation beyond isolated model deployments. SF Compute and Daytona fit the same thesis: tools that move AI from laboratory capability to real-world scale.