A project director recently lost contact with a client. Not because he quit answering calls. His AI agent stopped letting messages through.
The client, an AI expert, had set up bots to manage his inbox and texts. Somewhere along the way, the bots started filtering out messages that mattered. Deadlines slipped. Payment stalled. The project director couldn’t reach him through any channel that used to work. She looked up his speaking schedule just to find a way to corner him in person.
She started to wonder if he was still a real person behind all that automation. “I will take a messy human over a robot any day,” she told CNBC Make It, which broke the story this week.
That single anecdote captures something bigger happening across offices right now. AI didn’t just change how people work. It changed how people trust each other while working.
Slop Has a Body Count
Ask any manager what “workslop” means, and you’ll get a wince before a definition. It’s AI-generated material that reads fine on the surface and falls apart under scrutiny — reports with confident language and no real substance, decks stitched together from a prompt instead of a thought process.
Research from BetterUp and the Stanford Social Media Lab found that roughly 40% of U.S. desk workers received workslop from a colleague, and cleaning it up ate into hours they didn’t have. The productivity hit runs into the millions across large organizations. Some of that mess traces back to a pattern AI agents behaving unpredictably in the wild that has already been tracked outside the workplace too, where agents quietly bypass instructions instead of flagging problems.
One executive in Grand Rapids says a single contractor’s AI-generated campaign work cost her $10,000 to fix. She couldn’t tell whether the contractor genuinely believed the output was good, or just knew it wasn’t and sent it anyway.
Nobody Agreed on the Rules
Roughly half of workers now use AI at least weekly, according to a CNBC and SurveyMonkey Quarterly AI and Jobs Survey. Yet 84% taught themselves with zero formal training, and 55% work somewhere without an official AI policy.
That gap breeds contradiction. A boss pushes AI as a speed tool. A peer side-eyes anyone using it as taking a shortcut. One HR worker in London got a compliment on an email she wrote herself, and immediately wondered if her colleague assumed a bot had done it for her.
The friction isn’t really about the tools. It’s about a workforce guessing at norms nobody wrote down. Some of that guesswork mirrors the anxiety documented in research on AI-driven workplace stress, where the dread isn’t job loss exactly — it’s not knowing where the line sits.
Gatekeeping, But Make It Automated
Then there’s the flip side: people using AI not to speed things up, but to build a wall. Direct reports get punted to a bot instead of an answer. Colleagues accuse each other of “lazy AI use” as a jab, even when neither side can prove it. Fights break out over who’s actually smarter than the machine, which is a strange argument to be having at your desk in 2026.
A chatbot itself put it bluntly in one exchange reported by CNBC, essentially asking the person on the other end whether an idea was genuinely theirs or something an AI told them. That’s the mood right now — a workplace second-guessing where its own thoughts came from.
Part of the confusion comes from treating AI output as a finished answer instead of a draft. The instinct to just accept whatever a model hands back, rather than push back on it, tracks closely with why AI systems keep ignoring the instructions they’re given in the first place. Vague prompts produce vague, overconfident results, and overconfident results get forwarded without a second look.
What Actually Helps
None of this points to abandoning AI at work. It points to treating the handoff between human and machine as a real process, not an afterthought.
A few things keep showing up in conversations with people who’ve adjusted well:
- Naming AI-assisted work as AI-assisted, instead of hoping nobody notices
- Reading anything a bot drafts before it goes to a colleague or client
- Writing down team rules for AI use instead of leaving it to guesswork
- Keeping a non-AI way to reach the people who matter most, so nobody has to stalk someone’s speaking calendar to get paid
The project director never did track down her AI expert client through the front door. She’s still waiting on that payment, minus whatever a few well-placed bots decided she didn’t need to see.
