In Defense of AI Workslop: Nothing Exposes Useless Work Like Doing More of It

I remember attending a seminar years ago and learning about some fantastic KPIs I should keep track of. So first thing Monday morning, I declared a new edict that henceforth we would be tracking these metrics and promptly voluntold the team to start gathering the data.

As life would have it, a few months later I attended yet another seminar and discovered an even better set of KPIs. And what’s better than one dashboard the manager never looks at? Two dashboards the manager never reads.

Of course, by the time the weekly reports started landing in my inbox, I rarely had time to read them. After all, I was too busy discovering the next silver-bullet KPI and delegating the data collection and reporting for that one too.

Years later, I would stumble across some poor soul putting together one of these reports and ask, “Why are you wasting your time on this report? It’s useless!” Only to discover that I was the one who had required it to be created.

One emerging complaint about AI at work is that it produces mountains of useless stuff. More very long emails. More reports. More things for someone else to read — or, in my case, quietly ignore. And the blame usually lands either on the AI that produced the slop or the person who used AI to produce it.

But I increasingly think workslop is mostly a process problem masquerading as a people or technology problem.

The slop was there the whole time. The machine just runs it at full speed now, and scale has a wonderful way of making previously invisible waste and stupidity painfully obvious.

AI, after all, is exceptionally good at imitation. Give it enough examples of how your organization works and it can reproduce them with frightening efficiency. So when the machine perfectly imitates what your company has been doing for the last ten years and suddenly nobody likes the result, there is an awkward question hiding underneath: Did the machine do a bad job, or did it perfectly imitate bad work?

I think friction by way of legitimately slow execution was an accidental rate limiter on organizational stupidity. Want another report? Someone has to make it. Want another analysis? Someone loses an afternoon. The forty-hour workweek was not merely a labor constraint. In a strange way, it was also an accidental quality-control mechanism. There were only so many pointless things an organization could ask people to do before it ran out of people to do them.

Imagine your job is to produce the legendary TPS Report from Office Space. Every company has some version of it. Nobody is exactly sure why it exists anymore, but it exists, and therefore someone must produce it. You used to make one TPS Report a week. Then AI arrives and now you can make fifty.

Fantastic. Productivity is through the roof.

Soon everyone gets a personalized TPS Report. Your manager gets one. His manager gets one. Every department gets its own variation. By every measurable output metric, the organization has become wildly more productive and efficient at generating TPS Reports.

Except now there are so many TPS Reports piling into my inbox that I finally open one to figure out what the hell it is. And discover it is an incredibly stupid report.

The problem wasn’t that AI generated a bad TPS Report. The problem was that the TPS Report should never have been part of the weekly operating cadence of the organization.

That is why I think workslop is an annoying but essential diagnostic tool. Would any version of this report have been useful? Probably not. Does anyone actually make a decision from it? And if the manager only needs an answer to a particular question, why are we generating an entire weekly report instead of letting the manager query the company’s existing data directly?

Worse still, if the company’s IT infrastructure can’t answer a basic question without some poor soul manually assembling a spreadsheet every Friday, then the problem probably isn’t that we need a better report.

That is probably the first business process we need to fix.

I remember sitting in committee meetings when I was younger where asking, “What does done look like?” could feel almost sacrilegious. Everyone was busy debating solutions. Asking what actual outcome we were trying to produce somehow felt like an irritating distraction from all the excellent solutioning underway.

We were very busy solving a problem no one had bothered to define.

Which makes the current workslop problem feel strangely familiar. We ask for reports without asking what decision they support. We build processes without asking what outcome they are supposed to produce. We build features because we can build them, not because we have clearly defined the customer pain they solve.

Then AI comes along, lets us do all of it a hundred times faster, and we act surprised when we are buried in garbage. We asked a dumb question and got a very long load-bearing answer.

Busywork is easier to hide at human speed. A trickle of pointless work looks remarkably similar to productivity. But a flood of pointless AI workslop turns the volume all the way up until the waste becomes impossible to ignore.