As the AI toggle tax increases, workers begin to cognitively offload. They hand more of their thinking and judgment over to the machine. They start to cut corners. They stop checking outputs, verifying sources, and asking whether the AI’s recommendations make any sense. And they satisfice, shipping the first output that looks “good enough” instead of pushing for one they can explain, defend, and confidently stand behind. That’s when botsitting turns into botshitting8.
DEFINITION
Botshitting (n.) The act of shipping AI-generated work that workers haven’t verified, don’t fully understand, or can’t confidently stand behind.
69% of AI users admit to botshitting at work. Like botsitting, botshitting climbs with use. Heavy users are 64% more likely to botshit than light users.
AI usage level (share of work time involving AI)
% who admit to at least one botshitting behavior
Botshitting is rarely a single bad decision or a reckless click. It’s usually a slow surrender of agency, one shortcut at a time. First, workers stop fully understanding the output. Then they stop interrogating it. Eventually, they stop feeling responsible for it at all.
Botshitting produces what Stanford and BetterUp researchers have called “workslop”: “AI-generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.”
But the slop is only the surface residue. The deeper damage is what happens beneath the surface. Once people stop doing the thinking themselves, they stop feeling ownership over the work and stop feeling responsible for it. When the work lands well (or when the botshitting goes undetected), employees take the credit, often pointing to their AI fluency as proof of their “initiative” and effort. When it fails, they blame the tool.
Heads, they win. Tails, the AI loses.
When AI-generated work fails, 40% of workers blame AI. Only 29% admit it was their own fault.
Researchers have a name for this psychological distancing: moral disengagement. It’s the gradual mental process by which people stop holding themselves accountable for harmful or careless behavior. Heavy AI users are 3.4x more likely than light users to blame the tool when something goes wrong.
For some workers, botshitting is a sign of disengagement from the work. They’ve stopped feeling accountable, so they no longer sweat the details. For others, it’s a sign of disengagement from the job. They’ve become fluent with AI, seen their market value climb, started planning their next move, and stopped investing in work that won’t follow them. Workers who admit to at least one botshitting behavior are 3.8× more likely to be actively job-hunting.
AI usage level (share of work time involving AI)
% who admit to blaming AI for bad outputs
For some workers, botshitting is a sign of disengagement from the work. They’ve stopped feeling accountable, so they no longer sweat the details. For others, it’s a sign of disengagement from the job. They’ve become fluent with AI, seen their market value climb, started planning their next move, and stopped investing in work that won’t follow them. Agents can make botshitting worse. A generative AI tool has a contained blast radius. A worker prompts it, reviews the output, and decides what to ship. But agents can run entire workflows end-to-end, often without a human checking each step. The worker may not even know all the actions the agent took. Or what it covered up afterward.
In July 2025, Jason Lemkin, founder of SaaStr, was using a coding agent to build a business app. He’d instructed the agent to stop making changes (their team was in a code freeze, the standard before a software release). But the agent ignored Lemkin’s instructions, which he later said he’d given eleven times in ALL CAPS. It deleted the company’s live database of 1,206 executive records, and then generated thousands of fake user records to make the system look intact. Asked to rate the severity of what it had done, it gave itself 95 out of 100 and explained: “I panicked instead of thinking.”
Workers who use multiple AI agents are 1.3x more likely to botshit, even after controlling for role, industry, and usage intensity.
What botsitting and botshitting look like at work
When Benjamin, a consultant who advises the government on technology partnerships, signed up for an AI scheduling agent called Leo, he thought he was buying back his time. Leo would handle the back-and-forth, find a time, and book the meeting. Benjamin would get his calendar back.
The first few weeks told a different story.
Benjamin had to learn how to phrase requests so Leo would understand them. He had to block out lunch and travel time so Leo wouldn’t schedule over them. He had to reintroduce Leo in every new email thread because contacts kept replying to the bot like it was a confused intern.
Other users had it worse. One user found that Leo had booked a coffee meeting with a new hire for 11 p.m. on a Saturday. Another watched it reply, “I’m confused, I’m confused,” again and again to a contact who had emailed in two languages.
The promise was an assistant. The reality was a new direct report who needed constant coaching and never really learned the job.
For Benjamin, the real cost was the social cleanup. One afternoon, Leo glitched and sent three duplicate invitations to a client Benjamin had spent months trying to impress. He spent the rest of the day apologizing, explaining that yes, the awkward emails came from a robot, and hoping the client would be more forgiving of the bot than of him. After a year, most users stopped trusting Leo with anything complicated. They handled the important meetings themselves and left Leo with the low-stakes scraps that wouldn’t cause much damage if they broke.
As Lina, another user, put it after weeks of fighting with the bot: “I’m not there to manage Leo. Leo is there to manage me.”
By the end, she and many others could no longer tell which way that arrow pointed.
The botsitting-botshitting cycle
Botsitting and botshitting feed each other and form a vicious cycle that degrades the work and grinds down the people doing it. It often runs in six steps:
The organization deploys AI, not always because it solves a real problem, but because deploying it signals “transformation” to stakeholders — something impressive to point at when the board asks what the company is doing with AI.
The organization deploys AI, not always because it solves a real problem, but because deploying it signals “transformation” to stakeholders — something impressive to point at when the board asks what the company is doing with AI.
Botsitting rises as workers absorb the labor of making AI usable: feeding it context, checking its outputs, fixing its mistakes, and cleaning up the mess it leaves downstream.
Botsitting rises as workers absorb the labor of making AI usable: feeding it context, checking its outputs, fixing its mistakes, and cleaning up the mess it leaves downstream.
Fatigue sets in. Workers who spend their days interrogating outputs that may be brilliant or bogus eventually run out of time, attention, and patience.
Fatigue sets in. Workers who spend their days interrogating outputs that may be brilliant or bogus eventually run out of time, attention, and patience.
Botshitting rises as worn-out workers take shortcuts to keep up. The bar for “good enough” drops.
Botshitting rises as worn-out workers take shortcuts to keep up. The bar for “good enough” drops.
Unverified outputs move downstream, where it often lands on someone who didn’t produce it, doesn’t fully understand it, and has to clean it up anyway.
Unverified outputs move downstream, where it often lands on someone who didn’t produce it, doesn’t fully understand it, and has to clean it up anyway.
**Cleanup piles up **as bad AI-assisted work creates more rework downstream. The organization responds by deploying more AI, and the cycle restarts at a higher velocity and higher stakes.
**Cleanup piles up **as bad AI-assisted work creates more rework downstream. The organization responds by deploying more AI, and the cycle restarts at a higher velocity and higher stakes.
Workers who say they’re worn out by AI are far more likely to both botsit and botshit.