Sometimes I really am stuck on a design and don’t know what to do next. Maybe I’m sleepy (hard to focus), exhausted (hard to care), or just stuck (in uncharted-for-me territory). I could try the First Principles approach, but honestly it’s kind of a repetitive process for smaller units of work.

Normally when I am pair-programming with LLMs, I go back and forth over and over and over during the planning stage until we have an approach that I think Makes Sense and Sounds Good. This works great when I have a clear vision for what the next unit requirements are. My mind is able to translate the conceptual sketch into implementation code with little effort.

However, sometimes thats not the case, and in the planning stage I am more or less stuck at “shoot… man, idk”. This is where I have started conjuring up an old dark-triad trick I learned from some coworkers who were masters of Weaponized Incompetence.

Don’t ask people for help to figure out the Right Way to use their system. People are busy and will likely not invest the time to provide thoughtful guidance. Instead, publicly declare that you are going to use the Wrong Way. Watch as they scramble to redirect you to a better approach.

Obviously that comes at a real cost when working with other humans (don’t do it!), but it does bring up another point I am acutely aware of.

your ability to recall is not your ability to remember

Remembering is storage of information, Recalling is retrieving it. I used to think my memory was atrocious – until I learned that its actually middling – mostly I just need a lot of help being prompted to recall.

So all that is to say that when I can’t figure out what to do, I just tell the LLM to do whatever slop its naive approach is! Often, once I see a design implemented in the actual code diff, I can very clearly describe why it’s the Wrong Way. After that, I just describe the opposite of the current implementations characteristics and am much closer to what the Right Way looks like.

So to all the nay-sayers nay-saying LLM’s generate slop and are useless, I refute that claim! LLM’s do generate slop, but sometimes Slop is all You Need!