Building a More Effective Coach
No more answering a bunch of boring questions before you chat with the coach
First week for buildinpublic.coach: 6 users that weren’t me. (3 that were, in case you were wondering😂)
Of those 6, 1 answered the basic questions that I started with. Ok, I can take a hint. Having gone through it 3 times, I agree: it sucks.
So I changed my approach a bit: instead of trying to build a profile to begin with, I refactored the app to no longer use an agent. Instead, when I receive a message, I kick off another task that has a single goal: extract any information it can about the user or their business, and I store that in a vector database. While that kicks off, I continue the flow of the current chat. I start with a prompt that asks what questions should be answered in order to successfully respond to the user. This generates a list of questions that I then try to answer from the collected user data.
Something I learned: in order to prevent the AI from just making up an answer, give it the option of saying “I don’t know” if the answer can’t be found. Actually, it’s better to have it say “Unknown” when in a chat context, because otherwise the AI will start judging you based on the things it thinks you don’t know.
Next, I added in a piece that allows the AI to cite sources. I scraped a few websites that I believe are great sources for people building in public, and I want to use this in two ways: give the AI better information sources, and drive traffic to those sources.
I’m torn on the idea of scraping content, but I’ve decided that I’m ok with it based on three things:
First of all, I’m only scraping once (ok, I will probably do it a few times as I get better with it and then occasionally to update the stuff I’ve got), so I’m not going to be hammering sites with traffic. Secondly, I’m trying to get people to visit those sites. I’m not claiming credit for their work, I’m openly saying that this is one of my trusted sources, and you should check out their work. Finally, I’m planning on doing a revenue share based on link clicks. That’s one of my longer-term “secrets” that I’ve not shared as widely.
Goal for this week: 4 more users to get me to the first 10. And maybe 1 or 2 paid users.
Kicking off Buildspace Nights and Weekends Season 4
Time for N&W again! If you weren’t aware, I also did seasons 2 and 3. And I finished both, which I’m proud of. Especially since I definitely didn’t break any sort of records for users or revenue. Neither season was particularly great. It became a little painful to fill out those weekly updates when my users + revenue weren’t going up at all.
But I completed them. I refused to give up. One of the biggest parts of N&W: It’s not about who starts (most people get in), it’s about who is left by the end (about 20% completion rate).
By simply lasting 6 weeks, I’m top 20%. I’ve got that part down.
Now, it’s time to get to the top 1%, which is gonna be a lot harder.
But I’m here for it.
(As an aside, I dropped a video yesterday about N&W being better than Harvard. Did I do it that way hoping for clicks/views? Shamelessly, yes. But also, I do believe that this is a shift in education that is going the right way)
Content Corner
A couple of podcasts that stood out to me last week. First up, has anyone ever experienced writer’s block? If so, check out this episode of Unthinkable with Jay Acunzo:
Also, I really enjoyed this episode that Danny Miranda did with Sean Frank
Finally, here’s a video that I did with a bit more on how I’m building out BuildInPublic.Coach
Until next week!
~Leo