# How one CEO built his own team of AI agents > Source: > Published: 2026-10-21 06:43:32+00:00 Hi Optimists, Last month I asked whether a [VC fund could run itself](https://operationsoptimist.substack.com/p/can-a-vc-fund-run-itself-building). This month I was more interested in seeing an actual CEO running an actual team of AI agents alongside his startup team. So we chatted with Mārtiņš Lasmanis. [Mārtiņš Lasmanis](https://www.linkedin.com/in/martinslasmanis/) is a Latvian entrepreneur based in New York City and the co-founder and CEO of [Supliful](http://supliful.com). The platform helps entrepreneurs and brand builders launch and grow their own supplement, skincare and food brands, handling product sourcing, packaging and shipping. Supliful has a team of around 50 people and has raised about $3 million in funding. ## **He failed twice first** It all began with the first model a few years back, where he simply wanted to create a personal assistant. “Zero result,” he told me. “Literally nothing.” Spending time on X did not help, because everyone seems to have it, but you are stuck with a non-usable prototype. He dropped it for most of a year. Second attempt, August 2025, Claude Code and it kinda worked. He had it build a CRM for user research, connected to Typeform and Mailjet. But he could never get it to sit on top of Slack the way he wanted. When he was telling about the first two failed attempts, I think it gives some real perspective into why some things are just not working on the first try. ## **What changed on the third try was better tooling** OpenClaw changed everything. Mārtiņš had a habit of dropping a Slack note to himself every time he saw an X thread of building an AI automation he wanted to try. When OpenClaw came out, he connected it to chat, added repos and docs, dropped it as a Claude project, and started prompting hard: help me set this up, I don’t understand what this does, and so on. His chaotic bookmark library became a starting ground. ## **What does his setup look like** Fast forward a year, he now has has seven agents living in a mission control dashboard, everything in Discord, each with a name, a job and a channel: - **Max** builds external products - **Rex** builds the system itself - **Dash** runs personal brand and growth - **Atlas** does Supliful operations and the investor update - **Amber** does nothing but hunt for improvements, so he doesn’t have to sit on X looking for them - **Nova** designs - **Pixel** is his ten-year-old daughter’s tutor A cron job wakes the main agent five or six times a day, and it messages the others. Each one reads its own memory and a “soul document” he wrote himself. Here is how Mārtiņš describes soul document: “Imagine there’s a chef. The chef has a pan, a spatula, a mixer. He uses those tools to produce different things. He’s always the chef, he isn’t anything else. The prompts live in the tools. but the soul describes what a chef should do.” It all runs on DigitalOcean, Supabase, GitHub, Cloudflare and Tailscale, with markdown files in folders for as reference. Again, fairly impressive for a non-technical founder to build this. And even more impressive, that there is governance element to it. His agents are self-correcting in nature, proposing improvements on their own system. He has a main agent that scores them, and then Mārtiņš later approves, comments or asks for more research. If you want to see more evidence of it, his content on Linkedin is what the agents are mostly coming up with, like this [company update.](https://www.linkedin.com/feed/update/urn:li:activity:7518330010361946116/) ## **So how much does it cost?** About $3k/month in the first two months, and now down to $500/monthly. He’s running the Claude Max subscription and API on top. Two crash loops so far, where the system ate tokens in a circle until he noticed. One of them cost $800 overnight. Roughly a full working day, most days, now goes on the agents themselves. But his personal mental bandwidth is the one he is the most concerned about: **The hardest part is that your capacity to execute increases, but you are wound up out of your mind. Always on, always building, no time to reflect. And it hits your mental health.** **Because the models build you the house and put in the windows, but they can’t install the baseboards. And you need the baseboards. So being non-technical is not an excuse to skip them, you need to do the heavy-lifting for everything to work properly.** ## **The verdict** Turns out third time is the charm, and Mārtiņš managed to build something he uses every day. Seven agents, actionable artifacts and a self-improving system. There is plenty to be optimistic about here. But the part that stayed with me was how much of himself still goes into making it work. The agents have increased his capacity to execute. They have also given him a whole new team to manage almost every day. And when I asked how this connects to the rest of Supliful, he said there are no shared principles and no single memory layer yet. His personal setup has not yet found its way to integrate into the team. I suspect plenty of teams will recognize that gap. One person builds something impressive, but making it part of an existing human team is yet to be figured out. And that’s what I want to explore in the next issues of Operations Optimist. ### Suggested reading by Mārtiņš - **[Alex Finn](https://x.com/AlexFinn?)** on X about agentic use cases Thank you, Mārtiņš, for sharing your insights! And as always, thank you, dear reader. This newsletter is free and runs on your support. Every like, comment or share puts it in front of someone new, and I appreciate it so much. See you in your inbox, Diana