Hey good folks and happy Thursday!

What a week with Claude going out, and then fascinating study on idiosyncrasies in AI fiction. Fio had a good write-up on it on LinkedIn here.

This week, some thinking on learning loops inspired by Satya’s X post.

We’ve built an AI tool on top of Claude that is continuously scaling. Our skills function as deeply informative prompts that we can chain together. This also makes them easy to export to MD files and use across any other LLM if in the off chance Claude goes down. We’ve synced the tool with our project management systems and CMS so we can maintain a single source of content production truth without hopping back and forth between tools.

But, most of our learning is done in our editing phase––manually. Recurring or repetitive edits make their way into skill updates weekly. Product messaging or narrative chances do the same. But even here, AI can definitely do more of the work.

Next, we’re working with our tool to audit itself against its own first draft and the human-edited version we approve. It will find the differences, document them, create a list of recent gold standard content types to use to inform any future assets. And, monthly, it will tell us where what it has learned is not in alignment with a skill, so we can update that skill and make sure we are continuously getting better and better output from our AI content creation tooling.

This might certainly get us closer to a closed loop editorial system, but closer doesn’t mean much when the finish line is still so far away. We still have humans edit all assets before they go live. Many of those edits are timeboxed. We still have a writing agency that doesn’t use AI at all––and they write a ton of content for us, typically campaign assets, research or thought leadership. And, we still have an on-staff writer who interviews our customers,partners, and internal team members to publish guest content and use cases.

We are building our own AI content production IP, and also, still investing in our human IP, too. Both are critical. Both have their strengths. And all of it still needs edits.

I remain long on the career prospects of great writers and editors. And as a mother of young children, think a lot about how AI should be used in their education. I’ve come to believe it is like math, in which you must learn to show your work before you can use a calculator. For writing, too, kids must learn how to do it on their own, understand what good is, hone their voice and their style, and ideally get hands-on experience editing the writing of others. All of this before AI is introduced. Because AI can produce grammatically correct content, but it struggles to account for taste.

That’s all for this week!

A NOTE ON THIS ADVICE

You do you!

One content marketer’s best practices aren’t always right for another one, though I do try to distill out the main concepts and core practices I believe everyone can benefit from. That said, you must use good judgment when deciding whether to take advice given from folks on the internet. I am an expert, and this advice comes from my direct experience, but I am not smarter than you, and I have nothing to gain or lose because of what you do.

THINGS KEEPING ME CONTENT

• What I’m buying: Just reupped on our favorite body wash––Salt & Stone.

• What I’m reading: This research!

• What I’m eating: Boudin is back in my life in a huggeeee way, and my Southeast Texan heart is sooo happy!

Thank you so much for reading. Let me know what you think by replying to this email.

Very excited to be here with y’all.

Tracey