Management as AI Superpower | Ethan Mollick

By: Ethan Mollick Source: One Useful Thing Type: newsletter

Summary

Ethan Mollick makes the case that management — not prompting, not AI literacy — is the core skill for the agentic era. He introduces a delegation decision framework based on three variables (human baseline time, probability of AI success, and AI process time for evaluation) and backs it with OpenAI’s GDPval benchmark data showing AI matching or beating experts on professional tasks. His MBA students at UPenn accomplished dramatically more in 4 days with AI than pre-AI students did in a full semester, not because they were AI experts but because they already knew how to scope problems, define deliverables, and evaluate outputs.

Key Ideas Extracted

Knowledge entries created:

Delegation Decision Framework — Three-variable equation (Human Baseline Time × Probability of Success × AI Process Time); includes three levers for improvement, GDPval evidence, jagged frontier implication
Delegation Documentation as Agent Prompts — Cross-field delegation documentation inventory (PRDs, shot lists, Five Paragraph Orders); six common structural elements
Talent-to-Direction Scarcity Shift — AI inverts scarcity: talent abundant, direction scarce; management skills as competitive advantage

Ideas folded into entries above (not standalone):

Three levers for improving AI delegation → folded into Delegation Decision Framework (optimization paths for the three variables)
GDPval benchmark shift → evidence supporting Delegation Decision Framework
Jagged Frontier applied to delegation → implication within Delegation Decision Framework
Order-of-magnitude acceleration → supporting evidence across Talent-to-Direction Scarcity Shift and Delegation Decision Framework
Lowered cost of pivoting → connects to Parallel Prototyping for Clarity (existing entry)

Notes

Published Jan 27, 2026 on One Useful Thing (Substack). Free article (no paywall).
Ethan Mollick: Professor at Wharton, author of “Co-Intelligence,” widely cited AI-and-work researcher
References his own Jagged Frontier paper and GDPval discussion
Claude Code specifically mentioned — used it to generate a full 1980s adventure game with one prompt, deployed in three prompts total
Google Antigravity also mentioned as tool students used
Software development highlighted as first profession to feel the management shift (clearly verifiable outputs, mature tooling)
Strong connection to CHJ/Scaling Agents concepts — the delegation documentation inventory (PRDs, shot lists, Five Paragraph Orders) maps to the Mechanism Layer thesis
Cross-references: Relates to the Scaling People → Scaling Agents inbox item’s management-as-agent-management thesis; also connects to Marily Nika’s AI product sense (evaluation as core skill)

Raw Content

Validated via WebFetch against live source 2026-02-16. Content captured from inbox clip (source: clip, 2026-02-14). Full article publicly available at source URL.

Article Structure

Opening — UPenn MBA startup experiment: Students in an executive MBA class were challenged to create a startup from scratch in 4 days using Claude Code, Google Antigravity, ChatGPT, Claude, and Gemini. Author has 15 years of teaching entrepreneurship — estimates these students got an order of magnitude further than pre-AI students working a full semester. Prototypes had core features working, ideas were more diverse, analyses more insightful. Key enabler: lowered cost of pivoting meant students could explore multiple directions freely.

The delegation decision framework: Three factors from the Jagged Frontier of AI ability:

You don’t reliably know what AI will be good or bad at on complex tasks
AI is fast — produces in minutes what takes humans hours
AI is cheap relative to professional wages; doesn’t mind if you discard most outputs

Three variables for the delegation decision:

Human Baseline Time — how long the task takes you
Probability of Success — likelihood AI produces acceptable output
AI Process Time — time to request, wait, and evaluate

Tradeoff: you’re exchanging “doing the whole task” against “paying the overhead cost” of AI evaluation, possibly multiple times. A 1-hour task with 30-min evaluation only works if success probability is very high. A 10-hour task justifies several hours of working with AI.

GDPval evidence: OpenAI study pitting human experts against AI on diverse professional tasks. Expert tasks averaged 7 hours (Human Baseline Time). AI took minutes but evaluation took 1 hour (AI Process Time). With GPT-5.2, Probability of Success reached ~72% (tied or beat experts). Math: on a 7-hour task with 72% success and 1 hour evaluation, you save ~3 hours on average. Failed tasks cost more (wasted prompting + review time), but successes are much faster.

Three levers to improve the equation:

Better instructions — clear goals increase Probability of Success
Better evaluation and feedback — fewer iteration attempts needed
Cheaper evaluation — less time to assess quality

All three are improved by subject matter expertise — experts know what instructions to give, spot errors faster, correct more efficiently.

Delegation documentation as AI prompts: PRDs, shot lists, design intent documents, Five Paragraph Orders (Marines), engagement scopes — all solve the same problem (getting intent into action) and all work as AI prompts. They share common elements: what are we trying to accomplish and why? Where are the limits? What does “done” look like? What outputs do I need? What interim checkpoints? What quality checks before completion?

Management as the meta-skill: Software developers at AI labs note their jobs shifting from mostly programming to mostly managing AI agents. Coding matured first (clearly verifiable outputs), but other fields will follow. Management 101 skills transfer: explain what you need, give feedback, evaluate work. But the scarcity has inverted — talent is now abundant, knowing what to ask for is scarce.

Conclusion: The MBA students succeeded not because they were AI natives but because they already knew how to manage — scope problems, define deliverables, recognize quality. Management training was accidentally preparing them for exactly this moment.