My Personal Guidelines for Using AI Respectfully
I think it was a year ago that I got introduced to the concept of “Work Slop”. Work slop is AI-generated documents, emails, slide decks, and other artifacts that initially look reasonable, but when you dig in are full of hallucinations, logical gaps, or other issues that create significant work for reviewers and confusion for your colleagues.
I’ve had a couple of experiences with work slop personally, including a very recent experience where I created some work slop. I had used Gemini for research, vetted the sources it found, and then asked it to draft a slide deck based on an outline I created. What I didn’t do was fully validate the deck until I was presenting it and someone asked about a number. In the moment I had no idea what it was or why it was there. The number was accurate and properly cited. It was in my outline. And if only I had properly reviewed the deck I would have been prepared to answer the question, and also realized it fit better in a different spot in the narrative.
It was embarrassing.
After that experience I decided to give myself some guidelines around how I use AI effectively without creating work slop. These guidelines don’t cover coding, but instead focus on planning docs, emails, chat messages, and all the other day-to-day work I do. Also, they are my personal guidelines based on my experiences. I hope you find them helpful, but I’m not advocating that everyone should use them.
Prioritize Relationships
The first thing I think about when choosing whether and how to use AI is to evaluate the purpose of a task or a piece of work. Some things we do are primarily about communicating information. But most things we do at work aren’t just about data and sharing information, they’re also about building or maintaining trust and relationships with others you work with.
That statement may seem weird, but human beings are, generally, a relationship-driven species. And software development has long been an activity that happens in teams. Sometimes people reach out to me for purely factual information, like “what percent of your team’s time is focused on AI dev tools?”. But typically they’re asking for my interpretation, my advice, or even just an acknowledgment that I saw and appreciated their work.
When interactions are only about information sharing, I feel okay using AI to draft my response (although it’s often faster not to). I used AI to help me answer the question above, and cited the AI usage. But for interactions that are about relationships and trust, I limit my use of AI to just spelling and grammar improvements. And even then, I reject changes that make the final result not sound like me.
Prioritize the Reader
It is easy to see the benefits of AI to a creator. You can generate a dozen reasonably coherent blog posts in a few seconds. If your boss asks for a strategy deck, you can have a draft in minutes. AI can save hours of time as a creator. But communication isn’t just about the creator, it is about the reader too. When AI hallucinates or when its arguments are disjointed or illogical, it creates work for the reader to fact-check and parse overly complex AI-generated writing. I don’t want to make more work for my readers. That doesn’t seem respectful.
Because of this, I do my best to only use AI if it will benefit the reader at least as much as it helps me, the creator. Of course I use grammar and spelling checkers. I use generative AI to identify missing and incorrect words. And I often use generative AI to provide an overall critique of my writing, prompting with something like “You are <target audience>, please review this blog post for clarity and logical flow. Identify and explain errors but DO NOT rewrite it for me. You don’t have to be nice.” These uses of AI help me as a creator, but also improve the final product for the reader by finding issues I missed in my proofreading passes.
Prioritize Brevity
I’m a loquacious human who memorized a few hundred SAT words in high school and is still trying to get the most out of that time investment. One of the most consistent pieces of feedback I’ve received throughout my career is to be more concise in my communication (and use fewer parentheticals). I’m also a manager who has requested a 1-pager and received 8+ pages back on numerous occasions. I understand the importance of brevity in communication.
Because of this, I’m supportive of using AI to make a document or email shorter and simpler, but not to make something longer or more complex. I won’t let AI rewrite an entire email or even sections of a document for me. But I do use AI to help me shorten things, with prompts like “This is repetitive and too long, what can I do to tighten it up? Just give me notes, DO NOT rewrite.” Or maybe, “This needs to be a 1-pager, which bits are most helpful to the point I’m making and which bits can I cut?” And I treat its feedback the way I’d treat feedback for any other reviewer, ignoring the parts I don’t agree with and incorporating the ideas that improve the overall output.
Own It
The main thing I try to do with AI is be open about my use of AI and own the results. My blog posts have an AI use statement explaining how I used AI while creating them. I’m open about my use of AI with my coworkers. I’m happy to say something like, “Here are some titles Gemini came up with, what do you think? Can we combine some of these ideas together?” I want to be as open as I can about how I use AI and when and why I choose not to.
Owning it also means taking responsibility for what I sign my name to. If I create something, whether that’s a social media post, a design doc, a strategy deck, a video, or anything else, I’m responsible for the contents. I need to know the contents and be able to answer questions about it. If I send it to someone for review I need to be confident it is ready for review and accurately reflects my thoughts and understanding.
What Does This Look Like Day-To-Day
These self-imposed rules shape my day-to-day AI usage. In general, I don’t use AI for rough drafts of anything written. Personally, the process of writing the draft significantly improves my understanding of a topic and forces me to decide which bits are the highest priority. I use AI mostly for revision, often going back and forth a dozen or more times until I’m happy with the final product. And after that embarrassing recent incident, I make sure I’ve not only reviewed the material, but that I can speak to and go into detail on everything I produce.
But I’m curious, do you or your team have guidelines for AI usage?