How many times does a person need to learn the same lesson?
I mean, really? Does anyone know?
Don't hire me to be your marketer, folks!
But, a bit more seriously, let me tell you about WorkshopAI.app The idea was to leverage AI to provide users an MFA style workshop experience with distinct AI personalities bringing different expertise to the discussion that unfolds before you. My genre of choice is fiction so I focused on that. I wanted to setup AI powered "agents" each with distinct personalities, response styles, creative obsessions and foci. Then they would be placed in conversation with one another regarding whatever piece of short fiction the user submitted. Disagreements, insights, analysis, commentary, all the stuff. What I was able to build didn't fully meet those high expectations.
Imperfections
Ultimately, I wasn't and still am not all that happy with the distinctness of the "personalities." In a real workshop different people are varying degrees of helpful, varying degrees focused on their own personal aesthetic obsessions. Everything gets filtered through a real person and all their personality and knowledge. I wanted to emulate this at least by having the responses from any particular agent be halfway identifiable as them. That is, a consistent voice relatively distinct from the others.
In reality, they all mostly sounded the same. Each person's "focus" or "expertise", what each agent is particularly supposed to care about would shine through a little bit but also they would all often target the same issues in the piece. Sometimes they would each have something distinctly useful but a lot of the time there wasn't enough differentiation. Often they were praising or criticizing similar aspects of the piece in similar ways.
Sometimes Testing is a Good Idea
In this case what I mean is testing the concept, whatever it is, whatever the value proposition is… Because in this case, with WorkshopAI, I spent an unforgivable amount of time not deeply engaging with the actual content that was being output. I was focused on getting everything "working" yes, I had conversations, I had differently named personas etc. I moved to styling and similar concerns much too soon in the process. I should have had a minimal interface and tweaked and used and repeat, repeat, repeat. Then I could have seen that maybe what I wanted wasn't fully possible.
Before even fully validating the concept I was moving on to try to develop a novel solution to be able to workshop novel length texts. That was silly. It was me chasing the thrill of creating with AI coding instead of doing the slow work, the work that was needed in the moment.
But I'm One User
But what if other people liked it? Right? I got to a point where I figured I would make it a thing, have a free beta, and see what people thought. I really believed (and still believe) that the app offers a lot of value. The workshops aren't perfect but can still be very effective at highlighting strengths and weaknesses. Also, the three additional tools I added, a timeline checker, a tense shift checker, and a cliche finder. These are also done using AI and honestly they work rather well! Imperfect but well.
What's Next?
I'm leaving it running indefinitely. I'm leaving open beta open for the time being so that people can try it for free. I still think there's potential here. Of course, me telling anyone about it might help. I've learned that lesson but it was a slow process.