Thoughts on Clay's AI Writing Policy
Clay's AI Writing Policy went viral and of course, i read it.
It underscores many of the difficulties I have run into working on an Agentic based development team. Of course, there is no "tone" when it comes to writing python, but other artifacts, that humans read are very important.
One of the good points that this policy makes is that many people read what one person wrote. To write too much is to waste their time. In my work, I need to submit a status report on a weekly basis. The AI came up a great multi-page report, but fundamentally, it is way to long. It's great for historical purposes and to look back and investigate things, but it's horrible to submit to the CEO/CTO/COO for them to get an understanding. None of them have the time to read through a 5 page report on what the Development Teams did last week.
So to support this the report is divided into two parts, the longer report and a message to the internal message boards. In order to make the internal message board have only the most important things, it is limited to 3000 characters. Things that are inprogress but just starting can be eliminated as they may show up in the next week. The main focus is in What Got Done and Attention Needed. I have to spend time to curate the report each week, but the AI does a good job of finding all the items I need to review.
Another item is that the AI loves to use and create Acronyms. I have rules at every level, my configuration, the project configuration, and my team configuration to never create an Acronym, and always spell it out the first use of every post.
The last item I'll go over is that the AI loves to create run on lists using em-dashes, semicolons, and other techniques to create long sentences. So if I ask it to give me an answer in 3 sentances, it will make them very, very long sentences. So I have to explicitly tell it not to combine lists into run-on sentences.