What Is Botsitting? The Hidden Labor Behind the AI Productivity Paradox
- 2 hours ago
- 7 min read
A new word arrived in 2026 for a feeling knowledge workers already knew well: the sense that AI finishes the task faster, yet somehow the day fills up anyway. The word is botsitting — and naming it has done what good vocabulary always does. It made an invisible cost visible.
This article defines botsitting, explains why it happens, and reframes it through the discipline of Work Management: not as a personal productivity failure, but as a predictable symptom of work that was never designed for humans and AI to share.
Botsitting, Defined
Botsitting is the unmanaged work of making AI output usable — feeding the tool missing context, checking its results, debugging its mistakes, re-running prompts, and cleaning up confident-but-wrong answers.
The term was coined by the Work AI Institute at Glean in its Work AI Index 2026, a survey of 6,000 knowledge workers across the US, UK, and Australia. The headline numbers describe a paradox: roughly 87% of digital workers now use AI, about three-quarters say it makes them more productive, and they report saving around 11 hours a week through automation. Yet only about one in eight say their organization performs significantly better as a result.
Where do the individual gains go? A large share is consumed by botsitting itself — an average of 6.4 hours per week, close to a full working day. The time AI gives back with one hand, unmanaged supervision quietly takes with the other.
Why Botsitting Happens
Botsitting isn't a sign that workers are using AI badly. It's what happens when capable tools are dropped into undesigned workflows. Three patterns recur:
The invisible burden. The supervising labor — re-prompting, supplying context the system lacks, catching hallucinations — is real work, but it appears nowhere in any plan, budget, or metric. It is performed in the gaps between "official" tasks, which is exactly why no one accounts for it.
The reverse centaur. The promise of human-AI collaboration is the centaur: a person amplified by a machine. Botsitting inverts it. The human stops being the one served and becomes the assistant — managing the tool, smoothing its rough edges, and absorbing its failures. The augmentation runs backwards.
The vicious cycle. Because the cleanup is invisible to management, it reads as "low adoption" rather than as a design gap. Workers get told to use AI more, not to use it better — and the people doing the most botsitting are markedly more likely to start looking for another job.
When Botsitting Becomes "Botshitting"
There is a more dangerous failure waiting at the end of an exhausting botsitting loop. When the supervising labor goes untracked and unrewarded, people start cutting the corner that matters most: verification. They ship AI-generated work they haven't fully reviewed, don't entirely understand, and couldn't defend if challenged.
Glean's researchers named this botshitting, and reported that a striking majority of AI users admit to it. The mechanism is subtle: because AI output looks polished and self-assured, it invites the reader to skip the judgment step. Confident formatting substitutes for verified thinking — and the downstream risk lands on whoever trusted the work.
Botsitting exhausts people. Botshitting is what exhausted people do next. Both are symptoms of the same missing structure.
The Reframe: Botsitting Is an Architecture Problem
Here is where the conversation usually stalls — and where Work Management has something specific to add.
Most responses to botsitting are personal: prompt better, choose a better model, try harder to verify. But the labor isn't invisible because workers are careless. It's invisible because the handoff between the human and the AI was never designed. No one specified what the AI is responsible for, what the human is responsible for, what context the AI must be given, or how its output gets checked before it moves downstream. In the absence of that design, every individual improvises the missing structure in real time — and that improvisation is botsitting.
Organizations don't suffer this because they adopted AI. They suffer it because they layered AI on top of existing chaos. The Work Management Institute's first principle — Clarity Over Chaos — is the diagnosis: AI cannot supply clarity a workflow never had. It can only execute the existing structure faster, including the absence of one.
This is why the metric most companies watch is the wrong one. Counting seats, prompts, and usage measures activity, not value. On the Work Value Pyramid, usage sits at the bottom rung — Activities — while the thing leadership actually wants, Outcomes, sits at the top. A workforce can generate enormous AI activity while organizational outcomes stay flat. That gap has a name now, and it is spelled botsitting.
How to Design Botsitting Away
The way out is not less AI or more willpower. It is AI Workflow Architecture™ — designing how work flows across people and AI agents deliberately, so the supervising labor is absorbed into structure instead of paid for by individuals. Four moves do most of the work.
1. Govern AI in the flow of work, not at the model. AI Workflow Governance™ asks where AI has authority inside a specific workflow, through three controls that map directly onto the causes of botsitting:
Explicit Delegation — define what the AI is authorized to do and where a human must decide. This is what ends the reverse centaur: roles are assigned, not improvised.
Reference Alignment — give the AI the context it needs by design. Most botsitting is context-feeding done manually, one prompt at a time; giving a tool access to data is not the same as giving it context, and the difference is engineered, not willed.
Drift Detection — build the checkpoint that catches output diverging from intent before it ships. This is the structural antidote to botshitting.
2. Treat review as a designed stage, not a personal habit. The cleanup step belongs in the workflow as an explicit handoff with an owner — a review queue — not as something each person squeezes in invisibly. In the language of the 7 Workflow Architecture Standards, this is Explicit Handoffs and Exception Readiness made real.
3. Make the work defendable. Botshitting thrives where no one is accountable for the signal a piece of work sends. WMI's IDEAS Model assigns a Signal Owner to outputs, and the Decision Transparency standard requires that how a conclusion was reached is visible. Together they restore the one thing botshitting removes: the ability to explain and stand behind the work.
4. Measure outcomes, not adoption. Replace vanity usage metrics with Measurable Performance — cycle time, rework, quality, and whether the workflow actually produces better results. You cannot manage botsitting until you stop rewarding the activity that hides it.
Where AI is integrated this way, the pattern reverses: workers are far less likely to feel worn out by AI, far less likely to ship work they can't explain, and they spend measurably less of their time botsitting. The variable isn't the model. It's the architecture around it.
Key Takeaway
Botsitting is the real price of the AI productivity paradox — the human labor that quietly offsets the hours AI saves. It feels like a personal failing and gets managed like an adoption problem, but it is neither. It is the predictable cost of putting AI into workflows no one designed for human-AI collaboration.
The organizations pulling ahead aren't the ones using the most AI. They're the ones architecting the work around it — so the supervision becomes structure, and the structure produces outcomes instead of exhaustion.
Frequently Asked Questions
What is botsitting?
Botsitting is the unmanaged work of making AI output usable: feeding the tool missing context, checking its outputs, debugging its mistakes, re-running prompts, and correcting confident-but-wrong answers. The term was introduced in Glean's Work AI Index 2026, which found knowledge workers spend an average of 6.4 hours a week on it.
What is botshitting?
Botshitting is the escalation of botsitting: shipping AI-generated work that hasn't been properly reviewed, isn't fully understood, and couldn't be defended if questioned. It happens when the supervising labor of botsitting becomes so unrewarded that workers stop verifying outputs.
Why does AI create so much extra work?
Because AI is usually deployed into workflows that were never designed for human-AI collaboration. When no one defines what the AI owns, what the human owns, what context it needs, and how its output is checked, each worker improvises that missing structure manually — which is what botsitting is.
How do you reduce botsitting?
Stop treating it as a personal habit and design it out of the workflow: govern AI in the flow of work (explicit delegation, context by design, drift detection), make output review an owned stage rather than invisible cleanup, hold work to a defendable standard, and measure outcomes instead of usage.
Is botsitting the same as "babysitting AI"?
Informally, yes — botsitting is the named, measurable version of what people mean by babysitting AI. Naming it matters because it turns an invisible, untracked cost into something an organization can design around.
Stop botsitting. Start architecting.
The supervising labor behind AI doesn't disappear because a tool gets smarter — it disappears when the workflow around the tool is designed. The Work Management Institute defines the standards, governance, and credentials for that work.
Learn how AI Workflow Architecture™ designs human-and-AI work as one system
See the AI Workflow Governance™ controls for delegation, context, and drift
Build the capability through the Certified Workflow Architect™ (CWA™) credential, starting with the foundational CAWM™ certification
"Botsitting" and the supporting research are the work of the Work AI Institute at Glean (Work AI Index 2026). The structural framing above — workflow architecture, governance, and measurement — is defined and stewarded by the Work Management Institute™ (WMI™), advancing the discipline of modern work management through education, standards, and professional certifications.