The mystery artifact
Instead ofI made you a dashboard. Everything you asked about is in there somewhere.
TryOpen the second chart. Signups dropped by a third after the price change. Do we roll it back?
Please don't make the meeting read your AI output. Read it first, then bring us the point.
Play Snake while it thinks. The game pauses by itself when you switch tabs to check on your prompt.
Top score 0000
AI made writing 65 pages take a minute. Reading them together still costs the room an afternoon. The work doesn't disappear when you hit share. It moves from you to everyone else, and somebody still has to read it, find what matters and decide what happens next.
Someone bringspages of AI output to a meeting withpeople, and everyone reads along. It took a minute to make. It costs the room.
Change the numbers to match your last meeting. Assumes about two minutes of reading per page.
Here's a quick test. If what you're showing isn't worth more than the prompt you typed, send us the prompt instead. We can all run it.
None of this is anti-AI. Let it write the 65 pages, build the dashboard and prototype the app. Then do the part that's still yours. Read it, decide what matters and bring us that.
Same rule, other shapes. Here's what to bring instead.
Instead ofI made you a dashboard. Everything you asked about is in there somewhere.
TryOpen the second chart. Signups dropped by a third after the price change. Do we roll it back?
Instead ofI already built an app for this.
TryI built a prototype this afternoon. To ship it, we'd need login, a real database, an owner and about three weeks. Is it worth that?
Instead ofHere's what ChatGPT says:
followed by 400 words nobody asked for.
TryWe tried that in 2024 and it didn't stick. Support ran it, so they'll know why.
Instead ofHere are the AI notes from today's call.
Three pages the sender didn't read either.
TryTwo decisions today: we drop the annual plan, and Sam owns the migration. Full notes are linked.
Before you share anything an AI made, fill this in. If you can't, it isn't ready to share yet.
I'm proposing blank.
Because blank.
I need blank from you by blank.
Real read-alongs, from advice columns, forums and the news.
Deloitte Australia delivered a 237-page review to the Australian government. A law researcher spotted references to papers that don't exist and an invented quote from a court judgment. The corrected version disclosed that a generative AI tool helped write it.
Damage: a partial refund of the A$440,000 fee.
A reader asks her team for background on problems so she can find the root cause. One coworker keeps pasting ChatGPT answers into the thread instead. Her example: ask whether vandals are behind a hole in the llama fence and whether cameras would help, and he replies:
“ChatGPT says we can patch the hole with a fence repair kit.”
In the comments, someone else wrote a three-to-four-sentence answer in Slack. Her colleague said it was a lot to read, so he ran it through ChatGPT.
Damage: the question was about the vandals.
Eight people on a call, three of them actually talking. Two minutes after it ends, a polished AI summary lands in Slack. Three days later someone asks what was decided about the API. It was in the summary, on line 4.
Damage: nobody opened it the first time either.
An engineering manager's blog, 2026
A PM vibe-coded a feature, tens of thousands of lines of it, and demoed it to the CEO. Everyone agreed engineering would tidy it up and ship it. The tech lead expected a few hours of untangling. He found the code had rebuilt several of the product's core functions, some of them subtly wrong enough to break other parts of the system.
Damage: a codebase nobody can fully own.
A newly promoted project lead suspects a new hire is using AI to listen and answer questions live in meetings. The answers come out stilted and sometimes make no sense. Halfway through asking the forum how to bring it up, the lead admits AI helped write the post.
Damage: none. Just irony.
A junior engineer kept presenting ChatGPT lists as his own recommendations. When his manager finally checked one, the ideas were out of scope for the project. He had already spent about two weeks exploring them.
Damage: two weeks.
It isn't only anecdotes. In a Stanford Social Media Lab and BetterUp survey of 1,150 US workers, 40% said they had received AI “workslop” in the past month. That's the researchers' name for output that looks like finished work but doesn't move the task forward. More than half of the people who got it felt annoyed, and 42% saw the sender as less trustworthy afterward.
Sources: Fast Company and NewsChannel 9. Someone has even named the person who forwards AI output without reading it. They call it a meat proxy.
Max Tokens is sharing a 65-page AI doc, and the decision is buried somewhere inside. Press Just tell me the point to find it.
Pricing strategy: a comprehensive deep dive (v7_FINAL_final)
For your next screen-share. Tap a square when you hear it.