How a non-coder, solopreneur built a 33-agent AI system named after Greek goddesses, with a content pipeline that produces scored blog posts in under 3 minutes.
*One of my favorite things Iโve done just for my own entertainment was to have AI rewrite an email that was sent as professional, but as Samuel Jackson for me, lolโฆ I was crying ๐
Woo hoo! Letโs go! Iโve been playing around a bit with building custom agents in Notion. Would love to learn more about your process. Very impressive!
Love the creativity of making your system human-friendly. How do you handle your perfectionism though? I could never be satisfied just spending 30 minutes editing the draft!
Haha, well, in theory itโs 30 minutesโฆ but itโs usually longer.
I created my own CMS for my site (was on WordPress for 17 years, so I knew what I wanted) - and I used the TipTap editor for all the formatting (custom dividers, calls to action, FAQs, internal linking, etc.), so visually itโs easy. The text editing def takes longer.
p.s. love your bioโฆ You had me at Dumbledore ๐ Iโve started falling in love with data, thanks to AI, where I can convert the data into visuals if I want.
Thank you Kim! Glad the odd precision of โDumbledoreโ made you pause :-) Data can be fascinating indeed if you know what you are searching for, and with AI it has become much simpler!
I am still very new on Substack - 1.5 month. I can see from your response mentioning โTipTapโ I have a lot to learn ๐ .
So how do you decide what you publish on your site vs on substack? Iโm following you to find out!
I publish everything on my website first, then I publish the same to Substack. I have some content Iโm testing solely on the website, so weโll see how it goes!
The naming instinct is doing real architectural work here, not just decoration. When you give an agent a domain, a mythology, a brief โ you're forcing yourself to define what it knows and what it doesn't. That boundary is what makes the system work. I've been building the same thing from a different direction (mycology, not mythology โ mine are called hyphae), and the insight I keep landing on is yours: the human isn't replaced, they're relocated. You're still the one who approves, decides, edits. The goddesses handle what doesn't need you. That's the design. Writing a book about exactly this โ glad to have found your work.
I've actually had to create a little architecture chart for myself so that I can remember their names, and you've totally inspired me to go deeper with the origin stories for each of the goddesses.
The architecture chart is a good sign โ it means the system has grown complex enough to need one. That's usually when the origin stories start doing real load-bearing work rather than just being decorative. Once you know why a goddess holds her domain, the briefing writes itself.
I want to push back - gently - on the direction this is heading.
Five divisions, a director per division, an orchestrator above them, a pronunciation cheat sheet.
The system is beautiful.
It is also a lot of surface area for something that produces "a scored, verified blog post draft in under three minutes."
The risk with agent sprawl: you stop building a workflow and start building an org chart. Every new agent is another context to maintain, another failure mode, another thing to debug at 11pm.
My bias - fewer, sharper agents. A scalpel, not a pantheon.
Three agents that actually run beat thirty that look impressive in a diagram.
I donโt have 33 agents to ship a blog post - itโs across my entire business, and I donโt run them all constantly.
Example: one agent to run a performance audit on my site (site load time, etc.). I manually trigger that once a month. Some have become skills directly in Claude (AEO/SEO agent).
All that being said, Iโve already tightened this up since the initial build.
I took some great ideas from this. Thanks for such a comprehensive article. Iโm curious about Delphi itself. Is that just a database in Supabase? What does it look like?
as someone who knows kinda almost 0 about this stuff this is AMAZING. thank you for this and can't wait to read the rest in this series! it's so fascinating to me lol . i love the whole greek goddesses thing!
https://github.com/petersimmons1972/proving-ground -- Okay. I'm always terribly interested in folks who use AI Agent personalities. But I'm also of the opinion we should measure those things, too. See how they work. I am absolutely NOT here to take away from anything you've done. I am trying to learn from what we've done. If you choose, I would love to find out how your agents work for you. These are some benchmarks that I wrote to measure my own agents. You can even see one of mine included. My own pages includes an article about it. If you feel like running it and sharing it (publicly or privately) I would be very curious to see and hear more. Thanks.
Hi Peter,Absolutely on the measurement! SOโฆ with these agents, I still have some wiring to do. There are a bunch that I trigger, but mainly because I donโt need them on a regular basis. Ex: a performance auditor for my site - load times, etc. Iโll do a follow-up post because I have to โsetsโ of agents (for lack of a better word) that can run without me.I have a directory on my website (took it out of the nav while I was deciding what to do with it because I wasnโt updating it often enough). Itโs a directory of AI terms, all categorized.That now runs every week: one agent does the research, one writes the content, and one verifies it matches the format, voice, and SEO, then it publishes to the site.My goal is to remove myself as the bottleneck wherever I can.Thanks for sharing yours - Iโll definitely check it out this weekend!Much appreciated ๐
So freaking cool! I love the detail you shared here -- it gave me some ideas to improve my agents... and I looooove the Goddess entourage you've got going for you. Can't wait to read more!
I just added another feature to my Hub this weekend under โTycheโ - the Greek Goddess of fortune, chance, and prosperity - for my Business Command Center (i.e., finance!).
The naming thing is doing real work - agents with Greek goddess names probably get better prompts written for them than 'agent_7'. 30-45 minutes versus 3-4 hours is close to what I've seen but the Watch Me Build framing is exactly right because the setup is not stable yet.
My version fell apart twice before it held together for more than a week. What the demos don't show is what happens when Athena and Artemis disagree on quality criteria. I wrote about where my version broke: https://thoughts.jock.pl/p/when-ai-meets-reality-ep3
Your point about the quality criteriaโฆ thatโs the key ingredient for anything with AI (better planning, more context up front, before any deployment).
Thanks for sharing - will read where yours broke and hopefully save myself a little grief, lol.
I'm over here naming my Claude tabs 'good project stuff' and 'shitty project stuff' while you've got a whole pantheon running your business hahahaha.
I love all of it, Kim
Happy Friday.
HAHAHAโฆ I laughed out loud at that one, thanks.
I save the bland names (or sweary names) for the big corporations with no morals, lol.
I ran out of names for them, Kim
MF #1 - #2 and #3
ah ah ah ah ah
HAHAHAHAโฆ
*One of my favorite things Iโve done just for my own entertainment was to have AI rewrite an email that was sent as professional, but as Samuel Jackson for me, lolโฆ I was crying ๐
This is how I use AI most of the times too ๐
This cover animation is giving Hercules! ๐ฉท๐ฆฉ
Bookmarking this post!!
ha, thank you!!!
You and your post on animated covers! ๐ซถ๐ป
Yes! I click for the animated cover and stayed for the amazing content.
Thank you, Sandra ๐
Woo hoo! Letโs go! Iโve been playing around a bit with building custom agents in Notion. Would love to learn more about your process. Very impressive!
Thanks, Fran,
I think I'll do a video or two. I'm still figuring it all out, so I just want to make sure I don't make it more confusing, lol
Love the creativity of making your system human-friendly. How do you handle your perfectionism though? I could never be satisfied just spending 30 minutes editing the draft!
Haha, well, in theory itโs 30 minutesโฆ but itโs usually longer.
I created my own CMS for my site (was on WordPress for 17 years, so I knew what I wanted) - and I used the TipTap editor for all the formatting (custom dividers, calls to action, FAQs, internal linking, etc.), so visually itโs easy. The text editing def takes longer.
p.s. love your bioโฆ You had me at Dumbledore ๐ Iโve started falling in love with data, thanks to AI, where I can convert the data into visuals if I want.
Thank you Kim! Glad the odd precision of โDumbledoreโ made you pause :-) Data can be fascinating indeed if you know what you are searching for, and with AI it has become much simpler!
I am still very new on Substack - 1.5 month. I can see from your response mentioning โTipTapโ I have a lot to learn ๐ .
So how do you decide what you publish on your site vs on substack? Iโm following you to find out!
I publish everything on my website first, then I publish the same to Substack. I have some content Iโm testing solely on the website, so weโll see how it goes!
The naming instinct is doing real architectural work here, not just decoration. When you give an agent a domain, a mythology, a brief โ you're forcing yourself to define what it knows and what it doesn't. That boundary is what makes the system work. I've been building the same thing from a different direction (mycology, not mythology โ mine are called hyphae), and the insight I keep landing on is yours: the human isn't replaced, they're relocated. You're still the one who approves, decides, edits. The goddesses handle what doesn't need you. That's the design. Writing a book about exactly this โ glad to have found your work.
Thanks so much, Tim ๐
I've actually had to create a little architecture chart for myself so that I can remember their names, and you've totally inspired me to go deeper with the origin stories for each of the goddesses.
Thanks again!
The architecture chart is a good sign โ it means the system has grown complex enough to need one. That's usually when the origin stories start doing real load-bearing work rather than just being decorative. Once you know why a goddess holds her domain, the briefing writes itself.
Would love to see it when it's further along.
Thanks, Tim,
Happy to share! Iโll message you the screenshots. Theyโre in my personal Hub I built which is password protected.
That would be great โ always curious to see how others are structuring these things. Looking forward to it.
33 agents to ship a blog post.
I want to push back - gently - on the direction this is heading.
Five divisions, a director per division, an orchestrator above them, a pronunciation cheat sheet.
The system is beautiful.
It is also a lot of surface area for something that produces "a scored, verified blog post draft in under three minutes."
The risk with agent sprawl: you stop building a workflow and start building an org chart. Every new agent is another context to maintain, another failure mode, another thing to debug at 11pm.
My bias - fewer, sharper agents. A scalpel, not a pantheon.
Three agents that actually run beat thirty that look impressive in a diagram.
Thanks for pushing back ;-)
I donโt have 33 agents to ship a blog post - itโs across my entire business, and I donโt run them all constantly.
Example: one agent to run a performance audit on my site (site load time, etc.). I manually trigger that once a month. Some have become skills directly in Claude (AEO/SEO agent).
All that being said, Iโve already tightened this up since the initial build.
Itโs been a process ๐
Thank you for the clarification. That makes a lot more sense. Thank you!
Youโre welcome ๐
It also shows me Iโm due for an updated post with better clarification!
I love this so much. Now I need clever names for my stuff
Let me know what you come up with!
I had to create my own โorg chartโ for them all so I remember who does what, lol. Iโm getting there ๐คช
Iโm excited to read this. Mine are mostly after Russian authors but the Greek goddess move is strong!!
Oooh, I like the Russian authors theme, too.
I started with Metis for Claude (goddess of wisdom & strategy)- then went all in with the rest. *I only wanted women agents ๐
I love it!! Saved this to read this weekend!
I took some great ideas from this. Thanks for such a comprehensive article. Iโm curious about Delphi itself. Is that just a database in Supabase? What does it look like?
Thanks, Whitney,
Iโve created my own operating system (for lack of a better word). I call it โThe Hubโ (KimsHub, lol.. just me, no need to be super creative).
So itโs a web app (built with Claude Code), Supabase is the backend. I took a screenshot but I canโt upload it here.
Give me a week and Iโll get a new post & video up!
as someone who knows kinda almost 0 about this stuff this is AMAZING. thank you for this and can't wait to read the rest in this series! it's so fascinating to me lol . i love the whole greek goddesses thing!
Thanks, Caitlin ๐
I had to create a visual map for myself of all the goddesses because I keep forgetting who does what, lol..
https://github.com/petersimmons1972/proving-ground -- Okay. I'm always terribly interested in folks who use AI Agent personalities. But I'm also of the opinion we should measure those things, too. See how they work. I am absolutely NOT here to take away from anything you've done. I am trying to learn from what we've done. If you choose, I would love to find out how your agents work for you. These are some benchmarks that I wrote to measure my own agents. You can even see one of mine included. My own pages includes an article about it. If you feel like running it and sharing it (publicly or privately) I would be very curious to see and hear more. Thanks.
Hi Peter,Absolutely on the measurement! SOโฆ with these agents, I still have some wiring to do. There are a bunch that I trigger, but mainly because I donโt need them on a regular basis. Ex: a performance auditor for my site - load times, etc. Iโll do a follow-up post because I have to โsetsโ of agents (for lack of a better word) that can run without me.I have a directory on my website (took it out of the nav while I was deciding what to do with it because I wasnโt updating it often enough). Itโs a directory of AI terms, all categorized.That now runs every week: one agent does the research, one writes the content, and one verifies it matches the format, voice, and SEO, then it publishes to the site.My goal is to remove myself as the bottleneck wherever I can.Thanks for sharing yours - Iโll definitely check it out this weekend!Much appreciated ๐
33 agents and a pronunciation cheat sheet, this is the most committed build post I've read in a while. the messy middle section is what makes it real.
Thanks so much,
Hopefully the messy middle helps other people, too!
So great Kim! Canโt wait to see which goddess is invited to join next.
Thanks, Kristina,
I invited Tyche- the greek goddess of fortune, luck & prosperity. Sheโs command central for finance in my hub ๐
Itโs really fun when I see a need for a new agent and can name it after the right goddess!
I named my Jotform AI agent after Arabella from Thomas Hardyโs Jude the Obscure!
Thatโs fabulous! I love it! Itโs the simple pleasures, right?
Exactly โค๏ธ
So freaking cool! I love the detail you shared here -- it gave me some ideas to improve my agents... and I looooove the Goddess entourage you've got going for you. Can't wait to read more!
Thanks so much, Cara,
I just added another feature to my Hub this weekend under โTycheโ - the Greek Goddess of fortune, chance, and prosperity - for my Business Command Center (i.e., finance!).
Itโs fun letting this evolve organically.
The naming thing is doing real work - agents with Greek goddess names probably get better prompts written for them than 'agent_7'. 30-45 minutes versus 3-4 hours is close to what I've seen but the Watch Me Build framing is exactly right because the setup is not stable yet.
My version fell apart twice before it held together for more than a week. What the demos don't show is what happens when Athena and Artemis disagree on quality criteria. I wrote about where my version broke: https://thoughts.jock.pl/p/when-ai-meets-reality-ep3
Thanks, Pawel,
Yea, itโs definitely not stable (or finished).
Your point about the quality criteriaโฆ thatโs the key ingredient for anything with AI (better planning, more context up front, before any deployment).
Thanks for sharing - will read where yours broke and hopefully save myself a little grief, lol.