Matt Shumer, an AI startup founder, published a piece this week comparing the current moment in AI to February 2020. He draws the analogy to the "this seems overblown" phase right before everything changes. I remember distinctly watching the events of first reported infections out of China and thinking COVID would be no big deal, and then March rolled around and exponential growth reared its head. GPT-5.3 Codex and Claude Opus 4.6 dropped the same day on Thursday of last week, and Shumer says he is no longer needed for the technical work of his own job. He predicts 50% of entry-level white-collar jobs could be gone within one to five years. If you have not tried out any of the newest models, I encourage you to explore their functionality. If you are like me, you played with AI last Summer and thought, “huh, this is mildly useful” and then went back to your regular habits of doing most tasks manually. Back then, the models were considerably less functional than what was released last week. I asked Claude to explain the difference in model functionality that came out last week and got the following response:
“Imagine hiring an assistant who could previously read about 50 pages and remember them. Now that same assistant can read 3,000 pages and still recall specific details from page 12. That's the jump in context window and retention.”
I have been skeptical of AI equity valuations, and still am, but the underlying technology is not a bubble, even if the stocks are. This is the most deflationary technology in human history arriving in the middle of the most inflationary monetary system in human history. That tension has to resolve somewhere. When labor markets get disrupted at the scale Shumer is describing, the importance of holding a hard asset with a fixed supply becomes impossible to ignore.