Divya van Mahajan

Botsitting: The Work Draining AI Gains

ai-daily-brief-podcast

Overview

This episode of the AI Daily Brief podcast examines a 2026 report by Glean and the Work AI Institute (part of their Work AI Index 2026) on a newly named workplace phenomenon called “bot-sitting” — the hidden, largely invisible human labor required to make AI tools usable in enterprise contexts. The host argues that this concept is important for understanding why individual AI productivity gains are failing to translate into organizational performance improvements.

Source video URL: not available

Prerequisites

  • Familiarity with enterprise AI adoption concepts (productivity tools, copilots, agentic AI)
  • Basic understanding of the distinction between non-agentic AI (prompt-response tools) and agentic AI (autonomous, multi-step AI systems)
  • General awareness of the current landscape of AI tools (ChatGPT, Claude, Gemini, Microsoft Copilot, Codex, Claude Code)
  • Awareness of common organizational change management challenges

Main Points

The Productivity-Paradox Setup

  • 87% of digital workers now use AI at work; 75% report it makes them more productive, saving an average of 11 hours per week
  • Yet only 13% say their organization is performing significantly better as a result
  • The host’s view: individual productivity gains do not inherently translate to organizational gains without a deliberate mechanism to facilitate that transformation — regardless of bot-sitting

Defining Bot-Sitting

  • Bot-sitting is the work required to make AI usable: feeding it missing context, checking outputs, debugging mistakes, rerunning prompts, and cleaning up confident but incorrect answers
  • Workers spend an average of 6.4 hours per week bot-sitting, which substantially offsets the 11 hours saved
  • AI now automates approximately 27% of workers’ output, with workers expecting that to rise to 35%; 57% want AI to automate even more than they believe it ultimately will

How AI Time Is Actually Distributed

  • The Work AI Institute categorized AI time into three buckets:
    • Learning and building agents (reading, experimenting, building workflows): ~27%
    • Actively using AI (completing work with AI): ~36%
    • Bot-sitting: ~37%
  • Bot-sitting breakdown by weekly hours:
    • Feeding AI context: 2.3 hours
    • Supervising outputs: 2.2 hours
    • Debugging: 1.7 hours
    • Cleanup/tool switching: ~0.2 hours

Productive vs. Unproductive Bot-Sitting

  • Productive bot-sitting: verifying high-stakes outputs, iterating on prompts to improve quality, adding domain context the AI could not know
  • Unproductive bot-sitting: reloading the same context across multiple tools, comparing outputs across tools because the first wasn’t good enough, cleaning up AI-generated work
  • Workers using multiple AI tools are 35% more likely to report frequent bot-sitting — a phenomenon the report calls the “AI toggle tax”
  • The exhaustion multiplier: for every 10% more time spent feeding AI context, workers are 25% more likely to report feeling worn out

Bot-Sitting Escalating to “Bot-Sh**ting”

  • Bot-sh**ting: the phenomenon where bot-sitting fatigue causes workers to cognitively offload too much to AI — stopping verification, shipping the first plausible output, and surrendering judgment
  • Described as “a slow surrender of agency, one shortcut at a time”: workers first stop fully understanding outputs, then stop interrogating them, then stop feeling responsible for them
  • When AI-generated work fails, 40% of workers blame AI and only 29% admit personal fault — an example the report calls moral disengagement
  • Heavy AI users are 3.4 times more likely than light users to blame the tool when something goes wrong
  • The tools with the biggest reported productivity gains (ChatGPT at 67%, Claude at 59%) also had the highest rates of bot-sh**ting (71% and 92% admitting to it at least monthly, respectively) — “the smarter the tool, the sloppier the worker”

The Identified Cycle of Failure

The report describes a six-stage failure loop:

  1. Organization deploys AI
  2. Bot-sitting rises as workers absorb the labor of making AI usable
  3. Fatigue sets in as workers realize their job has become checking AI rather than doing work
  4. Bot-sh**ting emerges as fatigued workers take shortcuts
  5. Unverified outputs move upstream into organizational processes
  6. Bad AI-assisted work creates rework and cleanup downstream

What High AI Achievers Do Differently (Individual Level)

  • High AI achievers are defined as individuals who report AI improved both their productivity and the quality of their work
  • They are more selective about where they use AI: high achievers spend ~38% of AI time on core job tasks vs. ~48% for low achievers
  • They orient bot-sitting toward the productive: they are more than twice as likely to treat AI as a valuable teacher, using the oversight process to improve their own AI skills
  • They reinvest the time saved (“the AI dividend”) into new skills, not just more volume of the same work
  • The host notes this distinction may be temporary as agentic tools increasingly take over core tasks entirely

What High AI-Achieving Teams Do Differently (Team Level)

  • They keep human accountability intact even when treating AI as a teammate: when AI underperforms, they iterate (try other tools, add context) rather than abandoning AI or blindly accepting output
  • AI adoption spreads peer-to-peer more than top-down:
    • A leader using AI makes an average employee 2.4x more likely to adopt it
    • A direct teammate using AI makes them 3.2x more likely
    • A cross-functional teammate using AI makes them 5.6x more likely — because cross-functional workers design for real organizational messiness, not idealized processes
  • High-achieving managers delegate 32% more coordination work to AI, freeing time for coaching, developing, and inspiring people
  • Manager quality strongly predicts worker comfort with AI in high-stakes HR decisions: workers with good managers are roughly twice as likely to be comfortable with AI involvement in performance reviews, pay decisions, and terminations

What Transformative Organizations Do Differently (Organizational Level)

  • Transformative organizations: the 13% whose employees report AI has significantly improved organizational (not just individual) performance
  • They measure relevant metrics (quality of work, productivity, output, time saved) rather than vanity metrics
  • 71% of workers in transformative organizations have visibility into their own AI usage data vs. 40% in non-transformative ones — making AI a feedback tool, not just a surveillance tool
  • They treat governance as a living system: 93% of workers in transformative organizations say AI policy is reviewed regularly vs. 55% in non-transformative ones; 91% say the rationale is explained vs. 57%
  • Trust in the organization’s AI strategy: 93% in transformative organizations vs. 57% in non-transformative ones
  • They invest in people, not just tools: 84% of workers in transformative organizations say AI skills are formally rewarded vs. 48% in non-transformative ones; 90% say adequate training is provided vs. 52%
  • Treating AI strategy as primarily a vendor selection decision is identified as a hallmark of non-transformative organizations

Key Concepts

  • Bot-sitting: The hidden human labor required to make AI usable at work, including context feeding, output verification, debugging, and prompt iteration
  • Productive bot-sitting: Bot-sitting activities that add genuine value, such as verifying high-stakes outputs, improving prompts, or injecting unavailable domain knowledge
  • Unproductive bot-sitting: Bot-sitting activities that represent waste, such as reloading context across multiple tools or cleaning up incorrect AI outputs
  • Bot-sh**ting: The downstream failure mode where bot-sitting fatigue causes workers to stop verifying AI outputs and accept or pass along work they cannot explain or defend
  • AI toggle tax: The cumulative productivity cost of using multiple AI tools and rerunning the same prompts across them
  • Exhaustion multiplier: The observed relationship where each additional 10% of AI time spent feeding context increases reported burnout likelihood by 25%
  • Moral disengagement: The psychological process by which workers stop holding themselves accountable for AI-assisted work that goes wrong
  • AI dividend: The time (averaging 11 hours/week) saved through AI automation, which high achievers reinvest into skill development
  • High AI achievers: Individuals who report AI improved both their productivity and work quality, distinguished by selectivity, productive bot-sitting orientation, and skill reinvestment
  • Transformative organizations: Organizations (approximately 13% of those surveyed) whose employees report AI has significantly improved organizational — not just individual — performance and outcomes
  • Agentic AI / agentic work: AI systems that operate autonomously over multi-step tasks (e.g., Claude Code, Codex) rather than responding to single prompts
  • Work AI Index 2026: The annual research report produced by Glean and the Work AI Institute tracking AI adoption and impact in the workplace, based on a survey of ~6,000 workers conducted December 2025–January 2026

Summary

The episode examines the Work AI Index 2026 report from Glean and the Work AI Institute, which introduces the concept of “bot-sitting” to describe the significant and largely invisible human labor — averaging 6.4 hours per week — that workers expend making AI tools functional, including feeding context, supervising outputs, and debugging errors. The report argues this hidden labor substantially erodes the 11 hours per week that AI saves workers, helping explain why 87% of digital workers report personal productivity gains while only 13% report meaningful organizational improvement. The host accepts the importance of naming and understanding bot-sitting while arguing the individual-to-organizational translation problem is more fundamental than bot-sitting alone can explain. The report further identifies “bot-sh**ting” — a moral disengagement failure mode where bot-sitting fatigue leads workers to stop verifying AI outputs and offload accountability — and finds it is most prevalent among users of the most capable tools. The path forward, according to the report, requires building human infrastructure at three levels: individuals who are selective and skills-focused, teams that maintain accountability and spread adoption peer-to-peer, and organizations that invest in people, governance, transparency, and relevant metrics rather than treating AI transformation as a vendor procurement problem. The host concludes that as agentic AI adoption accelerates, the dynamics described in the report are likely to intensify rather than diminish, and that genuine AI transformation requires organizational change, not merely tool deployment.