---
id: "lib-2026-06-16-the-ai-skills-nobody-is-teaching-and-everyone"
type: "library"
title: "The AI Skills Nobody is Teaching (And Everyone Needs) | AI Expert Ethan Mollick"
lang: "EN"
org: "Unknown"
source_type: "Talk"
date: "2026-06-16"
week: "2026-W25"
url: "https://nosov.ai/library/lib-2026-06-16-the-ai-skills-nobody-is-teaching-and-everyone/"
original_url: "https://www.youtube.com/watch?v=9YMYVb1ASCg"
tier: "B"
tier_name: "Useful, limited"
tier_reason: "Wharton's Ethan Mollick in a long Simon Sinek interview, drawing on his own BCG field study and OpenAI's GDPval benchmark; informed expert opinion, but a conversation, not new evidence."
summary_basis: "transcript"
original_minutes: "59"
tags: ["enablement", "lecture"]
---

# The AI Skills Nobody is Teaching (And Everyone Needs) | AI Expert Ethan Mollick

Tier B (Useful, limited): Wharton's Ethan Mollick in a long Simon Sinek interview, drawing on his own BCG field study and OpenAI's GDPval benchmark; informed expert opinion, but a conversation, not new evidence.

Summary based on the full transcript.

## In short

Simon Sinek's "A Bit of Optimism" podcast with Ethan Mollick (June 2026). Mollick argues that AI rewards experience and taste, that junior staff are often worse at using it because they cannot judge outputs, and that the apprenticeship model that built expertise is breaking. His practical advice: pay for a top model, choose the strongest "thinking" model, and give it harder tasks.

## Key numbers

- 48% to 84%: share of GDPval tasks where the best AI ties or beats experts, a year ago versus at recording, as cited by Mollick
- 7 or 8 hours vs 15 minutes: average expert time versus AI time on GDPval tasks set by professionals with 14 years' experience
- 20 bucks a month: subscription to one of three leading models Mollick recommends as the first step

## What it reports

- Prompt-engineering tricks (roles, bribes, "think hard") no longer matter much with current models; instructing it as you would a person is enough.
- In Mollick's BCG study, junior employees looked like heavy adopters but were often worse at using AI because they could not evaluate output.
- GDPval: tasks created by professionals averaging 14 years' experience took experts 7 or 8 hours versus about 15 minutes for AI; blind expert graders rated AI as tying or beating humans 48% of the time a year earlier and about 84% at recording.
- He estimates roughly three times savings in effort and cost when hard tasks are delegated, even after checking time and some rejected outputs.
- The value of AI depends on how quickly an expert can evaluate its output, which makes expertise more important.
- Apprenticeship breaks when juniors and managers both prefer AI to each other.
- Organisations built around fixed processes (two-week sprints, PowerPoint output) cannot absorb 100x individual output without redesign.
- Writing tip: give AI a large sample of your writing, ask for a two-page style summary, then use it as custom instructions.
- Expects guilds such as lawyers and doctors to legislate human sign-off; coders lack that protection.

## How they got it

Conversational interview; claims refer to his research and published benchmarks without detail.

## Limits

- GDPval is an OpenAI benchmark; Mollick flags this but accepts it.
- No sources shown for the BCG junior-employee claim in the episode.
- Much of the hour is general discussion of art, education and voice.

## Why it matters for companies

Two points are actionable. Experience is an asset in AI use, so senior staff should be trained and involved rather than left to juniors. And junior development needs deliberate design, because the grunt work that trained people is the first to go to AI.

Original: https://www.youtube.com/watch?v=9YMYVb1ASCg