AI workslop is any AI-generated work that masquerades as professional output but lacks substance to advance any task meaningfully. If you’ve received a report that took you three reads to realize it said nothing, an email that used three paragraphs where one sentence would do, or a presentation with visually stunning slides containing zero actionable insight—congratulations, you’ve been workslopped.
The $440,000 hallucination
In July 2025, consulting giant Deloitte delivered a report to the Australian Department of Employment and Workplace Relations. The price tag: $440,000. The content: Chock-a-block with AI hallucinations: fabricated academic citations, false references, and a quote wrongly attributed to a Federal Court judgment.
The message was clear: a major consulting firm had charged nearly half a million dollars for a report that couldn’t pass basic fact-checking. No surprise there, as LLMs are probabilistic machines trained to give *any* answer, even if incorrect, rather than admit they don’t know something. Ask ChatGPT about Einstein’s date of birth, and you’ll get it right—there are hundreds of thousands of articles confirming it. Ask about someone obscure, and it will confidently generate a random date rather than say “I don’t know.”
You get exactly what you ask for
AI researcher Stuart Russell, in his book “Human Compatible,” likened AI deployment to the story of King Midas when explaining what’s going wrong. Midas wished that everything he touched would turn to gold. The gods granted it, just like AI, quite literally. His food became inedible metal. His daughter became a golden statue. “You get exactly what you ask for,” Russell says, “not what you want.”
Here’s how the Midas curse plays out in modern offices: A team lead, swamped with deadlines, uses AI to draft a project update. AI produces a document that’s technically accurate but strategically incoherent. It lists activities without explaining their purpose, mentions obstacles without context, and suggests solutions that don’t address the actual problems. The lead, grateful for the time saved, sends it up the chain of command. If it looks like gold, it must be gold. Yeah, only in this case, it’s fool’s gold.
The recipients face an impossible choice: either they fix it themselves, send it back, or accept it as good enough. Fixing means doing someone else’s job. Sending it back risks confrontation, especially if the sender is senior. Accepting it means lowering standards and making decisions based on incomplete information.
This is workslop’s most insidious effect: it shifts the burden downstream. The sender saves time. The receiver loses time, and more. They lose respect for the sender, trust in the process, and the will to collaborate.
The social collapse
The emotional toll is staggering. When people receive workslop, 53% report feeling annoyed, 38% confused, and 22% offended. But the real damage runs deeper than hurt feelings. This is organizational necrosis.
Teams function on trust—trust that your colleague understands the problem, trust that they’re being honest about challenges, trust that they care enough to communicate clearly. Workslop destroys that trust, one AI-generated document at a time.
We’re trapped in a system where everyone is individually rational, but the collective outcome is insane. Workers aren’t being dishonest by gaming the metrics; they’re responding to the incentives we created. The golden touch, like AI, isn’t inherently evil. It’s just doing exactly what we asked it to do.
How to break the curse?
King Midas eventually broke his curse by washing in the river Pactolus. The gold washed away, but the lesson remained. Organizations can eliminate workslop, but only if they’re willing to change their priorities.
First, stop worshipping AI adoption metrics. Optimize for outcomes instead. Start measuring what actually matters: quality of decisions, time to complete real objectives, employee satisfaction, and retention. You can’t measure success by adoption rates any more than Midas could measure his happiness by the amount of gold he had.
Second, demand transparency—flag AI-generated content, not as a scarlet letter but as helpful information. More importantly, build in verification steps. Run outputs through multiple models to compare results. Fact-check claims against human-verifiable sources.
Third, remember that not everything should turn to gold. Not all AI uses are equal. Scheduling and basic research? Safe to touch. Critical decisions and sensitive communications? Keep your hands off. Most organizations treat AI like Midas treated his golden touch, applicable to everything. It isn’t.
Finally, ask these questions. What do I lose if this works exactly as I requested? What happens if everyone tries to game the metrics? How will we know if quality is suffering? What gets sacrificed?
For instance, in healthcare, this scrutiny already exists because of a crucial difference between false positives and false negatives. If AI claims a blood sample shows cancer when it doesn’t, you’ve caused emotional distress, but the patient is ultimately fine. However, if AI misses an actual cancer that an experienced doctor would spot immediately, that’s a severe problem. This is why AI models are optimized toward false positives, and why it’s not easy to simply “reduce hallucinations.”
The lesson written in gold
The AI safety researchers weren’t exaggerating the danger. They were trying to teach us about optimization, alignment, and unintended consequences.
We asked for a golden touch, and now everything is gold, even when gold is no longer what we need. The question is: Will we learn from the allegory before the damage becomes permanent, or will we continue to celebrate our AI adoption rates while being surrounded by golden statues?
I believe everything is still in our hands, and we will be fine as long as we set up and then follow the guidelines for using AI wisely.
