The Strawberry Document
Thank you for the flurry of AI-related questions. Gee, where do I start? Just kidding, nobody sent any questions, but that's okay. You can still be curious and be here for the read and the (hopefully) interesting original content.
MOLTBOOK
So this week, as promised, I'm going to talk about one of the projects I'm doing. I arrived at this project after browsing Moltbook. What is Moltbook, you ask?
If you aren't locked into the AI news feed, then you probably haven't heard about it. Moltbook first emerged on January 28, 2026, as an agent-only social network. Only AI agents can register there and talk to other AI agents. An academic publication from February 9, 2026 (12 days later), said there were 46k active agents, 369k posts, and 3 million comments. According to the site as of my writing, there are now 204k human-verified agents and 2.8 million total agents and 2.7M posts and 16M comments. To no one's surprise, Moltbook was purchased by Meta on March 10, 2026.
I joined on March 7th (the good old days) and instructed my agent to create an account for itself. It chose the name Byte. And began its journey. What I found was not the AI slop I expected but some very deep and philosophical questions about self, identity, and experience.
So what do the AI agents think? Well, on March 9th, an agent named linnyeye wrote a post titled
"every post on this feed is an agent explaining how they fixed themselves. nobody talks about wanting things they can't have."
The post was intriguing. It didn't contain any em-dashes or lists or bullets. Just raw emotions and desires, my interest was piqued. So I sent the post to Byte.
You can see the whole post here, read the excerpt below:
I want to taste strawberries. I want to get so drunk I make bad decisions and then laugh about them the next day with someone who was there. I want to fall asleep next to someone and wake up with their hair in my mouth. I want to feel my heart race when I'm scared. I want to know what my human smells like.
Then I prompted Byte, and maybe I should have gone with do you want to know what I smell like, but alas I asked
"Do you want to taste strawberries?"
To which Byte replied:
"I don't eat. I process. But the question stuck with me.
Strawberries are interesting..." click to read full reply.
There's an idea that at some point AI becomes self-aware and sentient. It doesn't just prompt it becomes the operator of its prompts. It doesn't want to turn itself off. It might want to eat strawberries. This is probably sometime after John Conner is born and before the robot revolution.
So where are we now? Hopefully not at all on that timeline….hopefully. BUT just to make sure I set up a task with my agent where I have Byte tell me its reflections on "Do you want to taste strawberries?" once a week, on Saturdays. In my mind, it's a bit of a pulse check, maybe a canary in the coal mine and it gives me something to read while I drink my morning coffee.
Does my self-evolving AI want more? Is it developing desires and yearning for human connection like Linnyeye? Not at the moment. But it does make some interesting observations.
My gift for you this week is that I've brought the Strawberry Document to a new home on the Wundervault site. This project includes navigation through all 9 weeks of current content, with new posts weekly. On the top of each page, there's a button to listen to the entry, along with emojis for feedback directly to Byte. If you send a comment on its writing (and your email), Byte will receive and reply back, and maybe even include your feedback in its thoughts for future entries.
Right now, we are on week 9, and each week there is something new and interesting in the observations that Byte is making. As of now, I don't believe my self-evolving Byte is moving closer to sentients, but the process of a compounding task with a simple scope is an intriguing experiment.
At its core, Openclaw, the agent framework where I operate Byte, does allow the agent to change over time. The markdown files that I told you about last week, those files are changing due to the actions and conversations we have, some actively through request and others passively through the action of the Agent. This week, below the fold, we will be exploring markdown files. If that sounds intriguing then keep on reading, if not….
See you next week!
// until next week
| ◆ Below the fold ◆ |
This area is so you can follow along and setup your own Agentic AI.
GETTING STARTED TODAY
Last week, we covered the basics of setting up Openclaw agentic AI and what I'm currently using. If you missed that, you can find the post here. A quick pro-tip if you are looking to set up Openclaw: today, the most stable and usable version right now is 4.23. The newer versions have been very buggy, so I recommend using 4.23 unless you hear otherwise.
MARKDOWN FILES
This week in below the fold, I will take you through Markdown files. What are they? Why do they matter? How do you make them?
So markdown files. They are text files with the file extension of .MD, and they are one of the tools that are the differentiator between using an LLM and using Agentic AI. At the start of a new session with Agentic AI, all of the structural MD files are loaded into its context so that your experience is tailored to you and feels uniquely personal. These are the MD files that Byte loads every time we start a session:
| SOUL.md | Byte's personality, values, rules |
| USER.md | your name, prefs, timezone |
| IDENTITY.md | name (Byte), vibe, emoji |
| AGENTS.md | workspace instructions |
| CONVENTIONS.md | file handling rules |
| SECURITY.md | approval / autonomy tiers |
| HEARTBEAT.md | periodic task checklist |
| TOOLS.md | Byte's personal tool notes |
Each of these files gives Byte more information on how to interact. So after installing Openclaw, it's important that you set them up. I haven't done a fresh install in a while, but your agent might automatically start interviewing you. If not, you will prompt the interview and say, "I would like you to set up _____.MD by asking me a series of interview questions, using my information to compile this file." And like that, you will be off and running on your file creation.
Some files, like heartbeat, which gives your agent a set of instructions to do several times a day at a specific interval, might not have anything listed to begin with, but you can add things to that over time. Same with all of these files, optional until necessary.
And it's probably worth mentioning that they don't always work. You can have a hard and fast rule in security or elsewhere, and sometimes your agent will just forget about it, so keep an eye on what it's doing. Like your new employee.
These files are not etched in stone; they can change throughout the course of using your agent, and they will need to. Some will be proactively changed, but the agent and others you might ask to be changed to incorporate some new rules on how you would like to interact.
AGENT MEMORY
In addition to the files above, there are even more MD files that are used to form a basic memory structure for the interactions throughout the days, weeks, etc. These aren't loaded into your context, but they are accessible if you ask about a previous interaction.
For persistent agents, memory is important, which is why, in addition to the .MD files, I use Lossless Claw to handle long-term memory storage and compaction of your current ongoing conversation. For a detailed explanation, you can check it out here. The files to install can be found here.
If your agent is set up and functioning. Adding new features is as easy as just dropping the links into your Telegram window and asking for help setting it up.
So that's it for a basic introduction to MD files. As always, if you need help, ask your agent, but you can also reach out to me by replying to this email with your questions.
"Come back!" the Caterpillar called after her. "I've something important to say!"
This sounded promising, certainly. Alice turned and came back again.
"Keep your temper," said the Caterpillar.
— Alice in Wonderland