Author: IRPA AI Senior Analyst, Kieran Gilmurray
This article explores a growing failure mode in workplace AI: output that looks polished but is not decision ready. ‘Workslop’ happens when AI makes it cheap to produce plausible drafts, notes, and summaries, while the real effort is pushed downstream to others who must verify and fix them.
After reading this article you will understand why workslop is rising now, which workflows are most exposed, how to measure its cost properly, and what leaders should do in the next 30 days to reduce it without slowing useful adoption.
Why workslop is rising now
Google expanded Gemini across Docs, Sheets, Slides, and Drive to help people create faster, while Google Meet automatic note taking is now set to turn on by default for certain meetings unless admins opted out. OpenAI has also expanded ChatGPT Business connected app actions so AI can write into Microsoft and Google apps when administrators enable it. That means more AI output is not just being drafted. It is increasingly being inserted directly into core workflows and systems.
At the same time, organisations are under pressure to “show AI use.” The March 10 HBR IdeaCast episode on workslop frames the issue as a leadership and organisational pressure problem, not an employee morality problem. Reuters also reported on March 19 that AI tool use is being integrated into performance evaluations at Accenture, which is exactly the kind of incentive that can encourage quantity over judgement if leaders are not careful.
What workslop actually costs
The obvious cost is time. BetterUp’s survey findings suggest workers spend around two hours dealing with each workslop incident, and its model implies a monthly cost of $186 per employee, or more than $9 million a year for a 10,000 person company. Whether those exact numbers apply in every organisation will vary, but the mechanism is credible: the sender saves minutes while the receiver loses hours.
The less obvious cost is decision latency. A summary that lacks context, a deck full of plausible filler, or a meeting note with no actual decisions does not usually fail immediately. It fails later, when another person has to schedule a follow up, ask for clarifications, correct the record, or quietly redo the work. That is why workslop often hides inside apparently productive teams. The output count rises, but throughput does not.
There is also a trust cost. The HBR discussion makes clear that workslop erodes efficiency, trust, and morale. Stack Overflow’s February 2026 work on the developer AI trust gap points in the same direction: usage can be high while trust remains low. When colleagues stop trusting what comes over the wall, every artefact becomes suspect, and review time expands across the whole system.
Which workflows are most vulnerable
The highest risk workflows are the ones where AI can produce something that looks complete before anyone checks whether it is useful. That includes project updates, strategy decks, status reports, meeting notes, executive summaries, internal briefings, and first draft client emails.
Meeting notes are a good example because they now risk being produced passively at scale. Google’s admin guidance makes clear that automatic note taking can become default on for some meetings unless administrators change the setting. If those notes are not reviewed by a human who adds the real decisions, owners, and next steps, they can quickly become another stream of polished but low value artefacts.
Security and triage workflows are another clear warning case. The Linux Foundation announced $12.5 million in new grant funding to help maintainers deal with an “unprecedented influx” of automated security findings, and Daniel Stenberg said the curl bug bounty was ending after an explosion of AI slop reports. This is not a marketing problem or a writing problem. It is what happens when review capacity becomes the bottleneck.
Where AI genuinely helps
Workslop is not an argument against generative AI. It is an argument against unmanaged generative AI. The strongest gains tend to appear in bounded workflows where humans remain accountable, outputs are easy to verify, and the tool is clearly positioned as support rather than substitution.
That is why the current institutional response is so revealing. NIST and GSA’s March 18 announcement focuses on improving AI evaluation science in real workflows, covering performance, security, and functionality before systems are adopted more widely. The message is simple: evaluation in demos is not enough. Organisations need to know how systems behave where real work happens.
Why this is a governance problem, not just a training problem
Training matters, but configuration matters too. Google’s note taking default change and OpenAI’s workplace write actions both show that leaders cannot rely on individual judgement alone. If vendors make output easier to generate, and in some cases easier to publish into systems of record, governance has to become a design choice made at admin level, not just a behavioural request made in a training session.
This is also where regional compliance starts to matter. In the EU, AI literacy and governance expectations are already in force with further obligations approaching. In the UK, regulatory guidance is evolving. In the US federal context, the NIST and GSA collaboration signals a stronger push toward measurable evaluation and procurement discipline. Leaders therefore need to treat AI quality as an operating issue, not a future compliance task.
A 30 day leader plan to reduce workslop
Start by defining “decision ready” for your five highest volume AI assisted workflows. A note, deck, or summary should not be considered complete unless it includes source basis, real context, named owners, next actions, and anything else required for a colleague to act without guesswork.
Next, configure defaults consciously. Decide whether automatic note taking should be on, whether write actions into workplace apps should be enabled, and who can share agents or AI workflows more broadly. The goal is to prevent passive output inflation before it starts.
Then add a lightweight quality gate before redistribution. Anything forwarded onwards should answer a few basic questions: what is verified, what still needs checking, what decisions are being requested, and who is accountable for correctness. This creates friction in the right place.
Finally, change the metrics. Stop praising draft volume and start tracking correction loops, follow up messages caused by unclear outputs, time to decision, and incident counts. That is how you expose the hidden tax workslop creates.
What leaders should measure instead of “time saved”
Time saved is a weak metric on its own because it says nothing about where the time went or whether the system got better. A faster draft that creates five more clarifications is not productivity.
The better measures are rework time, decision throughput, downstream clarification volume, defect rates, and usage inside approved tools. For higher stakes workflows, also track whether outputs are source backed, whether a human owner reviewed them, and whether there is a reproducible trail for what was generated and why.
Conclusion
Workslop is not primarily a model quality story. It is an operating model story. When AI makes it cheap to generate plausible artefacts, organisations that do not define standards, ownership, and review gates will see output volume rise faster than trust and verification capacity.
The practical response is not to ban AI or shame employees. It is to build a minimum viable quality system: configure defaults, define decision ready outputs, measure rework instead of draft count, and create accountability for what gets shared onward. Organisations that do this will still move quickly. They will just stop mistaking volume for value.
About the author: Kieran Gilmurray
Kieran is a globally recognized authority on AI, automation, and digital transformation, having authored multiple influential books and hundreds of articles that have earned him prestigious accolades, including being named a Top 50 Global Thought Leader and Influencer on Generative AI in 2024, a Best LinkedIn Influencer for AI and Marketing, Top 50 Global Thought Leaders and Influencers on Manufacturing 2024, Top 14 people to follow in data and one of the World’s Top 200 Business and Technology Innovators.
CLICK HERE TO SCHEDULE AN ANALYST CHAT
Links:
Originally posted in the IRPA AI Network — Announcements & Updates