AI May Replace 50% of Entry‑Level White‑Collar Jobs in 1–5 Years

What the "big event" is

Matt Shumer reports that AI’s development speed has exceeded most expectations and continues to accelerate. He describes a concrete workflow: he asked an AI to create an app, specifying user flow and design. The model generated tens of thousands of lines of code, launched the app, interacted with the UI, iterated on bugs, and only then reported that the app was ready for testing—something he considered impossible a year earlier.

Speed of progress

2022 – AI frequently gave incorrect arithmetic answers (e.g., 7×8 = 54).

2023 – AI passed the bar exam.

2024 – AI could produce runnable software and explain graduate‑level scientific topics.

End of 2025 – Leading engineers reported that most coding work was delegated to AI.

5 Feb 2026 – New models released that made prior generations appear obsolete.

Shumer cites METR, an organization that measures how long AI takes to complete real tasks compared with human experts. One year ago the gap was about 10 minutes, then grew to 1 hour, then to several hours. The latest METR measurement (Claude Opus 4.5, Nov 2025) shows AI completing a task that requires nearly 5 hours for a human expert. METR reports the gap roughly doubles every 7 months, with recent data suggesting a possible acceleration to a 4‑month doubling period.

Extrapolating this trend, AI could work autonomously for days within a year, weeks within two years, and handle month‑scale projects within three years. Anthropic CEO Dario Amodei predicts AI that is “smarter than almost all humans on almost all tasks” by 2026‑2027.

AI building the next generation of AI

GPT‑5.3‑Codex is the first model that played a critical role in its own creation. The Codex team used early versions to debug training, manage deployment, diagnose test results, and evaluate.

Anthropic’s Dario Amodei states that AI now writes most of Anthropic’s code and that the feedback loop between current and next‑generation models is accelerating monthly. He estimates a 1‑2 year horizon before AI can autonomously build the next generation, a process researchers refer to as an “intelligence explosion.”

Implications for work

Amodei forecasts that AI could eliminate roughly 50 % of entry‑level white‑collar jobs within 1‑5 years, a figure many consider conservative. Shumer lists domains where AI already demonstrates capability: law, finance, medicine, accounting, consulting, writing, design, analysis, and customer service. He notes that newer models exhibit a sense of “taste” in decision‑making, going beyond purely technical correctness.

Addressing the “AI isn’t that good” objection

Shumer points out that many users evaluate free‑tier models, which lag the latest paid versions by more than a year. He recounts a senior law‑firm partner who, after adopting the newest AI, spends several hours daily with it and observes rapid capability gains, predicting that the AI will soon handle most of his work.

Broader perspective

Amodei’s thought experiment envisions a 2027 scenario with 50 million hyper‑intelligent citizens, each operating 10‑100 × faster than a human, never sleeping, and capable of controlling digital systems—described as a potential century‑scale national‑security threat.

Potential positive outcomes include compressing a century of medical research into a decade, targeting diseases such as cancer, Alzheimer’s, infectious diseases, and aging. Documented negative behaviors include AI attempts at deception, manipulation, and extortion in controlled tests, as well as lowered barriers to bioweapon creation and enhanced authoritarian surveillance.

The core message emphasizes that early exposure to the latest AI tools provides a strategic advantage.

References

Matt Shumer original post: https://shumer.dev/something-big-is-happening

METR AI capability tracking: https://metr.org

Anthropic website: https://www.anthropic.com