Three mistakes in three months from someone that built the course on how to validate AI outputs.

“I’m not sure what this part means” is not what you want to hear on a client call. I scanned the document and raised my eyebrows when I saw the offending sentence. This was sloppy AI writing that I had missed in my review. I fixed it immediately and looked to see what else I might have missed. Then a month later, feedback from another client that a strategy document was more complicated than it needed to be. I took a step back and realized that AI ever the “over-achiever” had introduced the issue. But it was me that failed to simplify. I cringed. This was the second time slop had gotten through my validation process. Then a few weeks later, a question from an instructional designer about constraints in a course brief that confused her. AI had twisted my directional feedback into hard rules that no longer made sense. And worse, I hadn’t noticed it in the review process. This was now a pattern worth paying attention to.

Good intentions and technical skill hadn’t protected me from sharing AI slop (and a piece of me cried). The way I was using AI was hindering collaboration, rather than making it easier, as was the promise of the technology. Was I a part of the problem rather than the solution?

Whether you’re talking with strangers on Reddit or co-workers in Slack, people have strong words for those who create AI slop (and they’re all negative). The assumption is that the person who shared the document is lazy. My own experience shows that it’s more complicated than that. Slop can make its way into the work we do even when we take the time to include context and goals in our prompts. It gets through even when we review its outputs carefully before we share it as work with our name on it.

HBR found that 53% of employees admit sending AI work they knew was unhelpful or low quality, at least sometimes (link).That is self-reported, and it tells me there are forces at play that go beyond personal responsibility.

Earlier this year I developed a course to address the ways AI slop affects collaboration at work. At the time I believed the solution was a checklist and a reminder that your name is attached to your work, whether or not you use AI. And because slop erodes the trust your colleagues have in you, you need to make sure what you’re sharing with stakeholders is what they need to move the work forward. Nothing more, nothing less. (this isn’t just me spitting facts, it’s actually backed by research). I still believe in personal responsibility and what I teach in the course. But I also know it’s an incomplete picture.

Since the course launched I’ve been able to observe my own failures and examine what causes them. The one that took me longest to see was how my own bias for detail amplified AI’s bias for complexity. When I first began presenting to executives I was given great advice: focus on the problem you’re solving and the recommendation. Do not include the details of how you got there, the options you considered, or your analysis. Be ready to answer questions about all of it, but keep it out of the deck. By default AI output does the opposite. It injects its reasoning and logs a history of its decisions. I appreciate the thoroughness, that’s my bias, which is why it’s easy for me to miss.

The other failure was easier to spot: my brain reached its limits. Brainstorming, drafting and editing in one sitting is more than I can handle, and none of my expertise knowing good looks like helps once my eyes have glazed over. As a business owner I’m often creating my own deadlines and urgency. But I now see I wasn’t factoring in the breaks my brain needs between thinking through an idea, documenting it, and polishing it. As a result I wasn’t doing my best work as an educator. The pressure to move fast to meet a time commitment won out. Now I’m adjusting how I schedule my own work. Anything I have to share gets three time blocks, spread over two or three days.

I’ll be honest. I think this is a hard problem. The solution doesn’t just require personal responsibility, but an awareness of how organizational and cultural norms can create a sense of urgency that overrides quality (fail fast, anyone?). And removing the friction that used to slow down our thinking is creating a surge in high-cognitive-load tasks. In between them we now have fewer low-cognition tasks, the ones that give our brains the time they need to recover. It will take some time for us to figure out how we, as people, adjust the way we organize our work, so that we can do high-quality work that respects what our brains can actually do.

Look, you’ll still encounter material at work that feels sloppy. Assume good intentions, not laziness. They likely reviewed their work, but didn’t see what you see. Take a moment to tell your colleague which parts of their work you can’t use or find confusing, rather than talking shit with your work bestie.

✌🏻 Mariena