Divya van Mahajan

Don't Blame AI for Workslop

ai-daily-brief-podcast

Don’t Blame AI for WorkSlop

Overview

This episode of the AI Daily Brief podcast/video channel examines the phenomenon of “work slop” — AI-generated content that appears polished but lacks substance — which gained widespread media attention following research published jointly by the Stanford Social Media Lab and BetterUp (a workforce training company). The speaker argues that work slop is fundamentally a human and organizational problem, not an AI capability problem, and that solving it requires restructuring incentives, eliminating unnecessary work, and investing meaningfully in employee AI literacy. The speaker’s name is not explicitly stated in the transcript.

Source video: URL not provided (published 2025-09-28 on the AI Daily Brief channel)

Prerequisites

  • Basic familiarity with generative AI tools (large language models, image generators, coding assistants such as Codex/GPT-5)
  • General awareness of workplace productivity concepts (input vs. output measurement, organizational incentives)
  • Familiarity with the discourse around AI adoption in enterprise settings
  • Optional: awareness of the MIT “95% AI failure rate” study referenced early in the talk

Main Points

1. Context: A Media Moment Around AI Underperformance

  • A widely circulated MIT study claimed 95% of AI efforts were failing; it was later scrutinized for being based on interviews with only 52 executives and a reading of earnings statements where absence of explicit revenue attribution was coded as failure.
  • This created a media appetite for AI underperformance narratives, into which the BetterUp/Stanford work slop research landed.
  • Headlines from HBR, Fortune, and CNBC all framed work slop as a productivity crisis caused by AI.
  • The speaker notes BetterUp has a commercial interest in the research findings, as the company sells solutions to the problem the research identifies — though the speaker does not dismiss the underlying conversation.

2. Defining Work Slop

  • BetterUp defines work slop as “AI-generated content that looks good but lacks substance” — slick slides, lengthy reports, over-tightened summaries, or code without context.
  • A separate academic paper from Northeastern University, Stony Brook University, and Meta identifies the textual fingerprints of slop as: verbosity, vagueness, repetition, and incoherence.
  • The speaker notes these characteristics are not exclusive to AI; humans produce them too.
  • The study from BetterUp surveyed 1,150 U.S. desk workers; 40% reported receiving work slop in the past month; estimated cost: ~$186/employee/month, or $9 million annually for a 10,000-person company.
  • The speaker accepts the general concept as real and important, regardless of methodological reservations about this specific study.

3. The Core Argument: Work Slop Is a Human and Organizational Problem

  • The speaker’s central thesis: AI models are capable enough to generate valuable output; when they do not, the cause is typically the context in which the person is deploying them, not the raw model capability.
  • Work slop proliferates where organizational incentives reward input visibility (how much work is seen to be done) rather than output quality (how efficiently goals are achieved).
  • AI reveals and amplifies a pre-existing dysfunction: people doing work primarily to be seen doing it, not to advance organizational goals.
  • As Ethan Mollick (professor) noted: “Make more PowerPoints as an organizational incentive is not going to work out well.” He warns work slop will become a way to shift blame from managers and workers onto AI.

4. AI Reveals Broken Incentives and Unnecessary Work

  • Two divergent uses of AI emerge from organizational incentives:
    • Input-focused organizations: AI produces vast quantities of documents, slides, and memos with no accountability for usefulness.
    • Outcome-focused organizations: AI is used to accomplish goals efficiently, which naturally produces less unnecessary output.
  • A LinkedIn scenario shared by Chris O’Dell illustrates the parallel: a student asked not to use AI on a pointless assignment, and an employee asked to use AI on pointless work — both expose that the assignment itself was the problem.
  • Work slop is also revealing that large amounts of assigned work was never necessary in the first place — legacy processes that were never re-examined, or pure “work theater.”

5. Organizational Remedies

The speaker outlines three structural fixes:

  1. Shift incentive structures from measuring inputs (volume of deliverables) to measuring outputs (goal achievement and efficiency).
  2. Eliminate fake work — legacy processes and busywork that exist because they always have, not because they serve the mission.
  3. Align leadership and teams around the new outcome-focused ways of working so expectations are consistent from the top down.

6. Human and Skill-Level Remedies

Even after organizational fixes, work slop will persist without investment in AI literacy:

  1. Model quality outputs explicitly — employees need concrete examples of what good vs. bad AI-assisted work looks like, including the harder case of work that looks good but isn’t.
  2. Create structured time and space for learning — organizations mandate AI use but rarely carve out dedicated practice time. The speaker cites a recurring finding from thousands of executive interviews: “I don’t have time to learn the tool that’s supposed to save me time.” Assigning AI use as homework on top of existing workloads does not work.
  3. Build a culture of AI editing and iteration — the first output is a draft, not a deliverable. The speaker describes cycling through many versions of a cover image as an example of the required mindset. Taking the default output without iteration is a primary driver of slop.

7. The Developer Analogy: Adaptation in Practice

  • Agentic coding tools became so capable in 2025 that the question for software engineers shifted from “can AI do this?” to “how do we manage the new challenges this creates?”
  • A Google Cloud study of 5,000 developers found increased code output and quality but also increased code instability — illustrating that new capabilities come with new challenges requiring active management, not passive acceptance.
  • An OpenAI researcher joked about forgetting how to program and simply pleading with Codex/GPT-5 — the speaker uses this as a light illustration of the adaptation underway, not a condemnation.
  • The speaker argues all knowledge workers will increasingly need a manager mindset: set goals, delegate to AI, review and iterate on outputs to ensure alignment with goals.

Key Concepts

  • Work slop: AI-generated content that appears polished and complete but lacks substance, accuracy, or genuine usefulness; creates the illusion of progress.
  • AI slop fingerprints: The textual characteristics — verbosity, vagueness, repetition, and incoherence — identified by academic research as markers of low-quality AI-generated writing.
  • Input vs. output measurement: A distinction in how organizations evaluate employee performance; input measurement rewards visible activity (volume of work produced), while output measurement rewards goal achievement and efficiency.
  • Work theater: Work performed primarily to be seen performing it, rather than to advance organizational or mission objectives.
  • Context engineering: The practice of providing AI models with sufficient context, framing, and instruction to elicit useful, high-quality outputs; broader than prompt engineering alone.
  • Agentic coding: AI-assisted software development in which the model operates with significant autonomy to generate, test, and iterate on code.
  • Agent readiness audit: A process (referenced as conducted by the speaker’s organization) that evaluates how prepared an organization’s leadership and teams are to effectively use AI agents.
  • BetterUp: A workforce training and employee support company that co-produced the work slop research with Stanford’s Social Media Lab; has a commercial interest in the findings.

Summary

The speaker argues that “work slop” — the proliferation of AI-generated content that looks substantial but delivers little value — is a genuine and costly phenomenon, but that blaming AI models for it is a misdiagnosis. The models are capable enough; the problem lies in the human and organizational systems into which they have been inserted. Where incentive structures reward visible activity over actual goal achievement, AI simply supercharges the production of performative, low-value work. Where assignments were already unnecessary, AI makes it faster and easier to complete them without forcing anyone to question whether they should exist. The speaker’s remedies operate at two levels: organizationally, leaders must restructure incentives to prioritize outcomes, eliminate legacy busywork, and align teams around new ways of working; at the individual level, organizations must invest seriously in employee AI literacy — modeling quality outputs, creating structured practice time, and fostering a culture of editing and iteration rather than accepting first-draft AI outputs as final deliverables. Work slop is framed as an inevitable but surmountable waypoint in the transition to AI-integrated work, one that will persist as long as the underlying dysfunctions of organizational life go unaddressed.