Overview
- Recent reporting on May 23–24 centers on Every CEO Dan Shipper’s account that intense automation has expanded, not shrunk, his staff after the company automated broadly and grew from about four people to more than 30.
- Shipper and analysts say language models need human-supplied problem frames—precise instructions, criteria and scope—because benchmarks often preload those judgments and hide the expert input required to reach high scores.
- Enterprises are creating new roles such as AI engineers, senior reviewers and domain translators to review AI outputs, fix thousands of small automation scripts, and manage ongoing token and orchestration costs.
- Large-scale investment and measured deployment diverge: OECD data show VC funneled 61% of global funding into AI in 2025 and higher in early 2026, while an Anthropic economics paper finds observed model use in firms is far below theoretical task coverage; Anthropic’s CEO has also warned entry-level white‑collar roles could shrink.
- The practical takeaway for businesses and workers is that durable value will likely come from tools and services that support human‑agent collaboration, quality control and maintenance, even as some routine entry‑level tasks contract.