esc to close

← ai impacts
issue 01 · ai impacts

2026 Q2 edition — "the productivity paradox and the institutional lag"

Apr 16, 2026 ·20 min read

2026 Q2 Edition — “The Productivity Paradox and the Institutional Lag”

A monthly analytical review of AI’s impact across software engineering, management, law, medicine, and science.

Preface

This is the first issue of a monthly series examining how artificial intelligence is materially changing the nature of professional work across disciplines. The series is designed for practitioners, researchers, executives, and decision-makers — people who need to understand not just what AI can do, but what the evidence actually says about what it is doing, at scale, right now.

This issue draws on research published through April 2026. Where data conflicts, we note it. Where findings are preliminary, we say so. The aim is not to inspire or alarm, but to inform.

In this issue: AI adoption is nearly universal across professional domains — but value is concentrating fast. 74% of AI’s economic gains are captured by just 20% of organizations. Experienced software developers are, in controlled trials, slower with AI tools than without. 90% of lawyers use AI daily, yet courts are increasingly sanctioning attorneys for filing AI-generated fictitious citations. Healthcare AI shows a persistent gap between controlled-study performance and real clinical deployment. And across all five domains — software engineering, management, law, medicine, and science — three patterns repeat: a verification gap, the erosion of entry-level development pipelines, and the concentration of gains among those already ahead.

Full analysis coming soon.

For the Lazy
The summary you actually wanted
01
The macro picture
74% of AI's economic value is captured by just 20% of organizations
The divide isn't who has AI — nearly everyone does. It's who reorganized around it.
19%
slower
Experienced developers in METR's RCT took 19% longer on tasks when using AI tools — not faster.
Lab result
55% faster
GitHub Copilot controlled experiments show this speed gain.
Real world
19% slower
METR's randomized trial of experienced devs on real codebases.
Legal AI
90% of lawyers use at least one AI tool daily
But courts are sanctioning attorneys for filing AI-generated fictitious citations.
Three patterns across all domains
The Verification Gap
Every domain's biggest failure is humans not verifying AI output — lawyers filing fake citations, devs shipping insecure code.
Entry-Level Erosion
AI is doing junior work, quietly eroding the pipeline through which senior expertise is built.
Concentration of Gains
AI amplifies existing advantages. It is not democratizing capability — it is compounding inequality.
The question is not whether to use AI
It's whether you can verify it.
The profession that first enforces rigorous AI output verification norms will define the standard for all others.