Workers are spending over 6 hours a week botsitting AI, fueling job frustration
Knowledge workers are increasingly spending hours each week “botsitting” AI tools—reviewing, correcting, and supervising their output—rather than doing the creative or relational parts of their jobs. Many report that while AI can accelerate certain tasks and sometimes boost individual efficiency, the net productivity gains for teams are modest or unclear once rework, debugging, and oversight are factored in. Commenters highlight growing resentment and alienation as employers mandate AI use, automate the most satisfying aspects of work, and treat oversight labor as invisible even while using it to justify headcount and cost cuts.
Botsitting and workplace frustration
- Many see “botsitting” as low-skill, low-satisfaction work: supervising LLMs, correcting errors, and triaging AI-generated slop from coworkers.
- A recurring complaint is coworkers and managers passing off unchecked AI output (e.g., PRDs, specs) as their own thinking, eroding trust and respect for the craft.
- Some feel AI is working “instead of” them, with humans reduced to babysitters/assistants to the model.
Debated productivity impact
- Reported individual gains range from modest (~20%) to dramatic (2–3x) for a small subset of power users, especially in coding/sysadmin tasks.
- Others argue real throughput is flat or negative once you include debugging, rework, and “prompt fiddling.”
- Links to observational studies (e.g., Faros) are cited; some interpret them as showing net gains, others as evidence of lower effective productivity due to quality issues.
- Distinction is drawn between “effort productivity” (same output with less effort) and “business productivity” (more valuable output per dollar).
Changing nature of knowledge work
- Several compare botsitting to factory work or Amazon warehouses: humans in the loop only where machines fall short.
- Some enjoy the new “manager of agents” role, focusing on architecture and direction; others hate being turned into code reviewers and project overseers.
- People note that models increasingly jump straight to “solutions,” making them worse as investigative assistants.
Tooling, workflows, and guardrails
- Power users stress careful setup: sandboxed agents, explicit guardrails, planning steps, and permission checks to veto bad actions.
- Others find the planning/agent cycle slow, context-switch heavy, and less fun than direct coding.
- There is concern about opaque model changes by vendors and a push toward eventual self-hosting for stability and cost control.
Job satisfaction, identity, and mental health
- Many report sharp drops in joy and meaning at work when AI automates the parts they liked (craft, problem-solving, relationships).
- Some accept AI and reorient toward ideation and shipping; others contemplate quitting tech for trades or retiring early.
- Alienation, loss of pride in skill, and fears of depression or even suicidality among displaced creatives are explicitly raised.
Management incentives and labor dynamics
- Commenters see management and investors as primary AI boosters, sometimes even mandating AI use under threat of firing.
- There is skepticism that token-based business models incentivize efficiency; some suspect subtle pressure to use more tokens.
- Broader worries appear about layoffs of “incompetent” or merely middling workers, degradation of quality of life, and the need for unions or new social arrangements as automation expands.