I turned my promotion work into a process I could repeat with AI

I talked with AI about promoting nohmitaina, saved the plan as a Procedure, and recorded what happened as an Episode. Using it for my own work showed me what to improve next.

I wanted to go beyond remembering

When I started building nohmitaina, I thought about the noh mainly in terms of semantic memory: organizing concepts from what I wrote so that AI could read them later. But I still couldn't clearly feel how that changed the way I worked with AI.

So in 0.3.0, I added procedural memory. I could save a way of working as a Procedure and use it again the next time I tackled that job. I wanted to try it on real work of my own. I chose promoting nohmitaina itself.

Turn a conversation into a reusable Procedure

I started by talking with AI about where to introduce the app, what we needed to research, and in what order to work. Find communities whose members might be interested. Check each community's posting rules. Keep a list of candidates, prepare copy for the destination, then review it myself before posting.

I saved that approach as a Procedure so it wouldn't end with the conversation. The steps included both work for AI and the points where I needed to review something or make a decision.

An English example of the full promotion Procedure: finding candidates, checking rules, preparing copy, getting approval, and recording the outcome. Select the image to open it at full size.

Research thoroughly once, then reuse the candidate list

The first run needed detailed research into possible places to post. But I didn't want to repeat all of that research from scratch every time.

So the process starts with a thorough search and a saved candidate list. On later runs, it reuses that list and researches any gaps. Posting rules can change, though. Even when reusing a candidate, we need to check the current rules before posting.

That includes whether AI-assisted copy is allowed or the author needs to write it themselves. Remembering a candidate and deciding whether it is appropriate to post there this time are separate steps.

The Research node in the English example. Its instructions distinguish the initial search from later research to fill gaps. Select the image to open it at full size.

Review the copy, then post it myself

I followed the process to check candidates and rules and prepare copy for each destination. For this run, I reviewed the final text and posted it manually. Here is the English post I published on X.

Submitting a post doesn't mean it reached anyone. Two Reddit posts displayed removal notices from Reddit's filters. Visibility and responses still need follow-up. I can't call the promotion a success at this point.

Record what happened as an Episode

Afterward, I asked AI to record an Episode: what we did, what happened, and what to carry into the next run. I wanted to be able to trace the candidate research, destinations, copy, results, and remaining checks later.

The record kept submission separate from confirmed visibility or responses. Next time, I can look at the Episode and the Procedure to work out what comes next, instead of reconstructing everything from memory.

The Episode from the actual work, shown in English. It records what was done, the results that still need checking, and observations for the next run. Select the image to open it at full size.

Using it showed me what to improve next

Looking back through Episodes, I noticed another problem. Promotion records and development records were mixed together in the same list, making it harder to follow the work I wanted to review. It would help to filter Episodes by area of work.

That's an improvement idea I found by using the app myself. Turn a conversation into a Procedure, carry it out, save an Episode, and look back to decide what to change. Use it myself, find what's missing, then improve it. That was the loop I wanted to try.