Something Big Is Happening

Original: https://shumer.dev/something-big-is-happening

Author/Source: Matt Shumer, AI startup founder and investor, February 2026

Key Ideas: - The gap between public perception of AI and current reality is "enormous and dangerous" -- most people are evaluating AI based on experiences from 2023-2024 that are no longer relevant - The author describes being personally displaced from technical work: he describes what he wants built in plain English, walks away for hours, and returns to find it completed better than he could have done it himself - The pace of improvement is concrete and measurable: METR tracks task duration AI can complete independently, showing doubling roughly every 4-7 months, with recent acceleration - GPT-5.3 Codex was the first model "instrumental in creating itself" -- AI is now contributing to its own development, creating a feedback loop - The free tier of AI tools is over a year behind paid versions; judging AI by free-tier ChatGPT is "like evaluating smartphones by using a flip phone" - Almost all knowledge work performed on a computer is exposed to disruption in the medium term -- not just specific skills but general cognitive work - Dario Amodei (Anthropic CEO) predicts AI will eliminate 50% of entry-level white-collar jobs within 1-5 years

Summary: Shumer writes a personal, urgent letter to non-technical friends and family about what he sees happening in AI. He draws an analogy to February 2020, when most people were not yet paying attention to COVID-19, arguing we are in the equivalent "this seems overblown" phase of something much bigger. His central evidence is autobiographical: after six years building an AI startup, he found that the February 5, 2026 release of GPT-5.3 Codex and Claude Opus 4.6 crossed a threshold where he is "no longer needed for the actual technical work of my job." He describes AI systems that build apps end-to-end, open and test them, iterate on their own, and only present finished products for review.

The piece emphasizes the recursive nature of current progress. He quotes OpenAI's documentation that GPT-5.3 Codex "was instrumental in creating itself," with early versions debugging its own training and deployment. Dario Amodei reports AI is writing "much of the code" at Anthropic, with the feedback loop between current and next-generation AI "gathering steam month by month." Shumer translates METR's empirical measurements of AI task completion duration -- doubling every 4-7 months -- into concrete predictions: AI working independently for days within a year, weeks within two, month-long projects within three. His practical advice: spend one hour per day experimenting with paid AI tools, push AI into actual work rather than treating it as a search engine, build financial resilience, and rethink career advice for children.

Relevance to Economics Research: While written for a general audience, this article captures the "practitioner shock" that is an important data point for economists studying technology adoption and labor market disruption. The METR measurements Shumer cites -- empirical tracking of AI task completion duration with consistent doubling times -- represent exactly the kind of quantitative evidence economists need for forecasting AI's labor market impact. The recursive self-improvement dynamic (AI helping build the next AI) is relevant to endogenous growth theory. The piece also illustrates the information asymmetry between AI-adjacent professionals and the broader workforce, which has implications for models of technology diffusion and skill premiums during transitions.