In 2026, the promise of artificial intelligence was simple: delegate the busywork, reclaim your time, and focus on high-impact strategy. Yet, for millions of remote workers, the reality looks drastically different. Instead of stepping back to admire their hyper-efficient workflows, professionals are spending their days micromanaging algorithms.
We have officially entered a new era of work, defined by a phenomenon researchers are calling the AI productivity paradox. Digital workers are adopting AI tools at record rates, but organizational output is flatlining. Why? Because the time saved on drafting emails or writing boilerplate code is being entirely consumed by a new, unrecognized form of manual labor: managing the AI itself.
To understand why this is happening, we have to look back over forty years to a foundational concept in human-factors engineering known as the ironies of automation. By applying this classic framework to the modern workplace, we can uncover exactly why your team is spending hours "botsitting," why unverified work is flooding your channels, and how to redesign your collaboration stack to finally realize the true ROI of artificial intelligence.
What Are the Ironies of Automation?
The ironies of automation refer to a human-factors paradox where automating routine tasks actively degrades human skill. Because machines handle the easy work, humans lose the daily practice required to solve the complex edge-cases that automation inevitably fails to process, making human intervention slower and more error-prone.
First introduced in 1983 by cognitive psychologist Lisanne Bainbridge, the ironies of automation were originally used to describe industrial control rooms and airplane autopilots. Bainbridge noticed a critical flaw in how engineers designed automated systems: they automated everything that was easy to automate, and left the human operator to handle whatever the machine couldn't do.
This creates two massive, ironic problems. First, the human operator is expected to monitor a system that operates flawlessly 99% of the time, which is a task humans are neurologically terrible at doing. Our attention drifts. Second, when the machine inevitably encounters an edge-case and fails, the human is suddenly thrust into a high-stakes crisis without the benefit of "warm-up" practice. Because the human hasn't been doing the routine work, their skills have atrophied. They have lost the intuitive "feel" of the system.
Fast forward to 2026, and the ironies of automation perfectly describe our relationship with generative AI. We use AI to generate reports, write code, and summarize meetings. The AI handles the easy 80%. But when the AI hallucinates a financial figure, misunderstands the nuance of a client's request, or generates code with a subtle security vulnerability, the human worker is forced to step in.
Because the worker didn't write the initial draft, they lack the structural context of the document. Finding and fixing the AI's complex mistake often takes longer than it would have taken to write the document from scratch. The ironies of automation dictate that AI doesn't eliminate work; it simply shifts the workload from creation to high-friction verification.
The "Botsitting" Tax: 6.4 Hours of Hidden Labor
Botsitting is the unrecognized manual labor required to make AI tools function effectively. This includes gathering context, writing precise prompts, reviewing outputs for hallucinations, debugging errors, and constantly switching between AI interfaces and primary workspaces.
The ironies of automation have manifested in 2026 as a massive hidden tax on corporate productivity. According to the groundbreaking Glean Work AI Index 2026, co-authored with Stanford and UC Berkeley researchers, white-collar workers now spend an average of 6.4 hours a week botsitting.
Think about the workflow of a typical product manager using a standalone AI tool. They leave their video call, open a new tab, and begin the arduous process of context-feeding. They copy notes from a document, paste them into the chat interface, summarize the emotional tone of the meeting that the transcript missed, and write a highly specific prompt. The AI generates a response. The product manager reads it, realizes the AI missed a key dependency, and writes a correction prompt.
This is not automation; this is delegation to a highly capable but completely amnesic intern. The context switching cost alone destroys deep work. You are no longer a creator; you are a babysitter for a language model. You are botsitting.
This perfectly explains the glaring ROI gap in the modern enterprise. The Glean data reveals that while 87% of digital workers use AI and claim it saves them roughly 11 hours a week, a dismal 13% say their organization is performing significantly better. The time saved on drafting is immediately reallocated to botsitting and context-gathering. The individual feels faster because the initial generation is instant, but the end-to-end organizational velocity remains stagnant.
The "Botshitting" Crisis and Social Loafing
Botshitting occurs when employees submit AI-generated work that they have not verified, do not fully understand, or cannot confidently defend. This happens because human-AI teams suffer from social loafing, where human workers subconsciously reduce their effort, assuming the AI has handled the heavy lifting.
If the ironies of automation explain why fixing AI errors is so difficult, the concept of "social loafing" explains why workers are increasingly choosing not to fix them at all. The Glean Work AI Index 2026 uncovered a staggering statistic: 69% of AI users admit to "botshitting." They are shipping code they haven't thoroughly checked and forwarding summaries they haven't fully read.
To understand this behavior, we have to look at Maximilien Ringelmann's famous 1913 rope-pulling experiment. Ringelmann discovered that when individuals pull a rope alone, they exert 100% of their effort. But when they pull as part of an eight-person team, they exert only 49% of their individual capacity. The larger the team, the less effort each individual applies, because accountability is diffused. This is known as the Ringelmann Effect, or social loafing.
In 2026, we have inadvertently introduced social loafing into single-player workflows by treating AI as a teammate. When a human and an AI co-author a document, the human subconsciously assumes the AI "pulled the rope." The human assumes the AI checked the math, verified the syntax, and maintained the brand voice. The human skims the output, slaps their name on it, and ships it.
This is the dark side of a poorly implemented AI copilot for teams. Because the ironies of automation make verifying complex AI output tedious and difficult, and social loafing reduces the human's motivation to exert effort, the natural result is a flood of unverified, mediocre work. The AI produces a highly plausible hallucination, the botsitting human fails to catch it, and the error gets embedded in the company's codebase or client deliverables.
Goodhart's Law and the AI Productivity Paradox
Goodhart's Law states that when a measure becomes a target, it ceases to be a good measure. By tracking AI adoption through vanity metrics like "prompts per week" or "percentage of code generated," companies incentivize employees to use AI performatively, generating useless volume rather than actual value.
Despite the rampant botsitting and botshitting, enterprise leadership continues to push for higher AI adoption rates. But how they measure that adoption is actively making the problem worse. Goodhart's Law is in full effect across the corporate landscape.
When executives mandate that 50% of all marketing copy must be AI-generated, or when IT dashboards track the number of tokens processed per employee, workers adapt. They start using AI to summarize emails that are only three sentences long. They ask the AI to rewrite perfectly good paragraphs just to hit their usage quota. AI prompts have become the 2026 equivalent of the mouse-jiggler—a way to simulate productivity without actually producing anything of value.
The data backs up this widespread enablement illusion. A May 2026 survey by Gartner's HR Practice found that 19% of employees report zero time saved with AI, despite having access to premium enterprise tools. Furthermore, a separate Gartner survey revealed that only 36% of leaders are very confident in their ability to redesign roles around AI.
We are trying to achieve an AI productivity payback by forcing a fundamentally flawed workflow onto our teams. Tracking the volume of AI usage completely ignores the ironies of automation. It doesn't matter how many lines of code the AI generates if the human developer has to spend four hours botsitting the output to ensure it doesn't break the production environment. Volume is not velocity.
Overcoming the Ironies of Automation with Contextual AI
To resolve the ironies of automation, teams must move away from standalone AI chatbots that require manual context-feeding. Instead, AI must be integrated natively into the collaborative workspaces where teams already operate, allowing the AI to automatically see, hear, and understand the shared context.
The root cause of botsitting is the gap between where work happens and where the AI lives. As long as your AI is in a separate tab, you will be forced to act as the manual bridge between your team's context and the algorithm's input field. To break this cycle, we have to eliminate the context gap entirely.
This is exactly why we built Coommit. We realized that traditional video conferencing tools and standalone collaboration apps were exacerbating the AI productivity paradox. If you have a meeting in Zoom, draw a diagram in Miro, and then open ChatGPT to summarize the action items, you are guaranteeing hours of botsitting.
Coommit solves this by combining high-definition video, an interactive real-time canvas, and built-in contextual AI into a single platform. When you use Coommit, the AI doesn't need to be spoon-fed context. It lives inside the meeting. It hears the nuance of the conversation, and critically, it sees the diagrams, sticky notes, and wireframes your team is manipulating on the AI whiteboard for teams.
Because the AI shares the exact same real-time context as the human participants, the ironies of automation are drastically reduced. You aren't asking the AI to generate work in a vacuum; you are asking it to synthesize and execute based on the shared reality of the canvas. This eliminates the 6.4 hours of botsitting required to translate human ideas into machine prompts. It also curbs botshitting, because the AI's outputs are instantly visible on the shared canvas, allowing the entire team to verify and iterate on the work collectively, rather than relying on a single, socially loafing individual.
Conclusion
The transition to AI-augmented work was never going to be as simple as flipping a switch. As the ironies of automation prove, handing the easy tasks to an algorithm doesn't necessarily make human jobs easier; it often makes them more complex, isolating, and tedious. The 2026 AI productivity paradox is a direct result of forcing teams to spend their time botsitting isolated tools and managing the fallout of unverified botshitting.
To truly unlock the potential of your remote workforce, you have to stop treating AI as an external vendor that requires a highly detailed brief for every task. You must bring the AI into the room. By unifying your video calls, your collaborative canvas, and your contextual AI into one seamless environment, you can finally stop managing your tools and start doing your best work. If you're ready to leave the botsitting behind, it's time to see how Coommit is redefining team collaboration.