Partttttyyy People!
We, like everyone else have been leaning into the agent world! We’ve been building Sol, our internal agent, and I went into the process assuming the hardest part would be the technical stuff, and booyyyyy was I wrong. 
The harder problems have been figuring out what it should actually do, getting three people looking at the same problem the same way, deciding what information she should have access to, and figuring out the human layers. We wrote a blog post going in depth: ZURB - You Don't Finish an Agent
So I’m curious about the “messy middle” for everyone else? If you’re building an agent right now, what’s breaking?
Technical, organizational, not knowing what do to or where to turn?! haha, I want to hear it all.
1 Like
Here’s what we’re doing next to push this forward:
Five things, in order:
- Get the transcripts somewhere Sol can reach without handing over the whole account. Dropbox is a blocker, we have to figure out: do we move to something custom? Do we separate the accounts? The workflow has to support the human element.
- Define the content templates so Ben has something concrete to built towards and work backwards off of
- Move the topic synthesis out of a chat window and onto a page that updates on its own.
- Make the health of the pipeline visible to people who aren’t Ben, so I can see what’s working without asking him. This also creates an opportunity for an agent to also help maintain this
- Define who’s we’re actually talking to, and how Sol, the agent, supports this effort.
Thoughts?
1 Like