When low-effort AI work gets passed around a company, does it flow sideways, upward, or to whoever has to act on it?

This explores where AI-generated workslop tends to land in an organization: sideways to peers, up to managers, or down to direct reports. To be direct, this collection has no study that tracks those directions. What it does have is research on where AI output gets made, how it moves, and who ends up checking it. Taken together, that research suggests workslop flows downstream, toward whoever has to act on a piece of work, more than it flows along the org chart.

Start with where it comes from. Workers hand off AI tasks mostly in information-heavy jobs, and the pattern follows what the technology can do rather than how often people chat with it Where have workers actually delegated tasks to AI?. Telemetry from heavy users adds a twist. Their activity in documents and spreadsheets went up by about 21%, while their messaging and meetings went up by only about 7% Does generative AI shift knowledge workers away from communication?. So people are producing more material alone and talking about it less. That is the setup for workslop: a polished document gets passed along without the conversation that would have shown how thin it is.

Next, who catches it. Research on multi-agent AI workflows finds that failures often land on people who neither wrote the original prompt nor watched the work happen Who actually bears the risk when multi-agent workflows fail?. That describes an AI pipeline, but the logic carries over to people. In a handoff chain, the person who pays for low-quality work is usually the one furthest from where it was made. Meanwhile, experts are being turned into custodians whose job is to check and manage AI output rather than create work themselves Does AI reshape expert work into knowledge management?. That points to one likely direction: toward the senior or specialist reviewer who signs off on things.

Why does workslop get through at all? One paper argues that people treat AI-generated material with full trust by default, partly because it sounds confident How much should we trust AI-generated data in inference?. Another finds that AI only boosts output when people apply skills they already have. When they work outside their expertise, quality and learning both suffer When does AI actually boost worker productivity?. Put those together and the most workslop should come from people working beyond their own skills, aimed at people who are inclined to trust it. That is a matter of skill gaps, not seniority.

The surprising part is that the work process itself may give workslop away. Wholesale AI handoffs show up as bursts of finished text that break from a person's normal rhythm of writing or coding, while genuine back-and-forth with AI leaves no clear mark Can process data distinguish AI delegation from ordinary collaboration?. Organizations may end up tracing workslop by when and how a document was made, not by who sent it to whom. To answer the original question with real numbers, this collection would need survey or workflow data on who receives what, and it doesn't have that yet.