TB: Easy Is Overrated (Newport)

An AI Task Force at Organization Science analyzed manuscript submissions before and after ChatGPT's release and found that heavy AI assistance increased the volume of submissions while measurably degrading their quality — readability dropped, jargon and nominalizations rose, desk rejection rates climbed, and acceptance rates fell. Newport's claim is general: tools that make a step easier are not automatically tools that make the work better, and the act of writing is itself part of how thinking happens.

• Submissions to the journal show a sharp post-2023 shift: papers with minimal AI use fell from ~100% of submissions to about 30%; papers with heavy AI use rose correspondingly.
• Reading-ease scores for submitted manuscripts fell 1.28 standard deviations between January 2021 and January 2026.
• Heavy-AI manuscripts exhibit "longer words, more complex sentence structures, more jargon, and more nominalizations" — the polish that looks like quality is the opposite of it on measures that predict whether readers parse and absorb prose.
• Desk-rejection rates: ~70% for heavy-AI papers vs. ~44% for papers without AI.
• Acceptance rates: ~3.2% for heavy-AI papers vs. ~12% for low-AI papers.
• The editorial decisions were made without knowledge of AI use, so the correlation is not editor bias against AI — it is decision-makers responding to the writing itself.
• The same pattern shows up in hiring departments seeing similar quality collapse in AI-generated applications.
• Newport's higher-order claim: doing writing by hand develops capacities — disambiguation, structure, taste — that the tool was supposed to replace; "easy" routes around those capacities and so they fail to form.

• "The manuscripts are arriving in greater volume … the writing feels weightless in a way that rarely describes academic writing." — Organization Science AI Task Force
• "Most people assume that AI produces cleaner, more polished text … but on the measures that capture whether a reader can actually parse and absorb the prose, AI writing is worse."
• "Making things faster or easier is not the same as making things better."
• "These tools make individual researchers' lives easier in the moment … but they are leading to worse outcomes for the field as a whole."

A rare quantitative result on the cost side of AI-assisted writing. The collective-action structure of the finding — easier for the individual, worse for the field — is the same shape as the agentic-PR anti-pattern Willison flags: cost is offloaded onto downstream readers. The piece is also a counterweight to John's own pro-AI stance from Tyler Cowen on AI and Academia; it does not deny the productivity gains, it specifies where the bill comes due.

• Agentic Engineering (Willison) — the "delegating to reviewers" anti-pattern, applied to academic readers
• Paul Graham on Good Writing — why effortful prose is a thinking tool, not just a communication tool
• Tyler Cowen on AI and Academia — the other side: the cost of not using AI in academia
• The Beauty of Slow Research Assistants (Fletcher) — the attention-bottleneck argument that pairs with this

Easy Is Overrated — Cal Newport, calnewport.com, 2026-05-11