A recent UI revamp of my site meant moving the "how to's" from the top of each feature page, to a centralized Documentation section. This was an intentional design decision, not something forced upon the site because of the new UI. As I get more and more comfortable with Claude Code, I decided to let it write the documentation.
My first attempt wasn't great. I just told it to scan the site and create the documentation. Never having written documentation myself, or holding a technical writer position, I didn't craft a very good prompt. I was unimpressed with AI's result.
My second attempt involved using multiple agents (Agent Teams) and specifically asking for several different agent types: lead (default behavior for Claude's Agent Teams), a technical writer, a UI designer, and an industry/domain expert in the music business, working together to create the documentation. The prompt gave more context for each role than just naming them. I'm just trying to limit the length of this post. ![]()
The result was incredible to watch as I asked Claude to complete one section for me to review (I wanted to see if it was worthwhile continuing). Claude started by spinning up the industry expert agent, doing research that would specifically capture how musical artists and their management teams use today's tools and platforms for all things related to artist management. That research then fed the technical writer agent, guiding it on why this feature on my site is valuable to the intended audience (artists, management). The technical writer was also asked to create charts, tables, call outs, mini how to's, etc., improving the usefulness of the documentation itself. Then the UI designer took the work of the domain expert and technical writer, built the UI and fed in the copy. It took a few minutes but what I saw when it completed was amazing. Claude was ready to continue with the rest of the sections and I quickly said, "YES!".
The lead agent then proceeded to start up several domain experts and technical writers, each paired and taking on a specific section. The lead agent decided it was best to work on two sections at a time. When the 2 agent team was complete, I'd see the UI designer agent fire up and start building the section. The whole process took roughly 20 minutes and I'm extremely happy with the result. Yes, I'm having to read, review, and revise a little here and there, but I'll gladly do this.
Moving from a single agent, whether it be Google AI Studio (where I started) or Claude Code's default single agent approach, to Agent Teams (still considered beta by Anthropic) is yet another "game changing" event in AI history.
Happy to answer any questions about my process or set up.