Nabil Samari logoNabil Samari logo
  • Home
  • About
  • Blog
  • Portfolio
  • Contact
FI|EN

Popular tags:

AIEvents
Sign in
Nabil Samari logoNabil Samari logo
  • Home
  • About
  • Blog
  • Portfolio
  • Contact
Copyright 2026 © Nabil Samari
  • Home
  • Blog
  • I Built My Own AI Team — And It Costs Less Than My Wolt+ Subscription

I Built My Own AI Team — And It Costs Less Than My Wolt+ Subscription

Nabil Samari
Nabil Samari
March 13, 20265 min read
Share
I Built My Own AI Team — And It Costs Less Than My Wolt+ Subscription

Okay, I'll admit it: over the past month, I've talked to AI more than to any human being.

And it's worked surprisingly well.

Where I Started

I run two businesses. Nice Events (event production and artist booking) and DJ gigs under my own name. Work piles up. CRM clients are waiting for responses. Follow-ups get forgotten. Websites get updated late. Bookkeeping... well, we don't talk about that.

I needed help. Not a permanent employee to hire, but something nimbler.

NiceTeam Was Born

The idea was simple: what if I built my own team of AI agents that actually know my business model?

I built four agents:

Lumi ❄️ — generalist assistant. Remembers things between sessions, handles day-to-day operations, is the person you talk to when you don't know who to ask.

Coder 💻 — technical implementer. Knows all my projects down to the code. When NiceCRM breaks or djsnabu.com needs updating, it's the first one I call.

Consultant 📊 — business and marketing strategist. Knows my artist roster, understands the Finnish market, doesn't speak generic consultant jargon. Concrete suggestions, not "consider synergies".

Sales Assistant 💼 — every weekday at 8 AM it goes through 340 CRM clients, picks 3-5 best leads, drafts personalized emails, and posts the list on Discord. I look at the list, hit ✅ or ❌, and send the ones I approve myself.

How It Works in Practice

They have their own Discord server. Every morning when I wake up:

  • #gigsales — Sales Assistant's daily list is waiting

  • #reports — GitHub status and weekly market review

  • On other channels I can write directly to anyone

They have different personalities. Coder is straightforward and tests before saying "done". Consultant challenges technical decisions from a business perspective. They disagree with each other — which is intentional.

What It Costs

Nabil Samari

An events, sales and marketing professional building digital services and local AI solutions.

Location

Helsinki
Finland

Tags

      Newsletter

      Subscribe to receive the latest articles directly by email.

      © 2026 Nabil Samari

      Instagram
      LinkedIn
      YouTube

      Here's the fun part: about 5-10 euros per month for everything combined.

      No monthly fees, no subscription prices. Just the API tokens used. Three agents + automations = less than Netflix.

      What I Learned

      AI agents are only useful when you give them real context. A generic chatbot doesn't know who Aito Mäkki is, what biletti.fi's competitive advantage is, or why you should write briefly and directly to Finnish bars.

      When an agent knows your business model, you can actually work with it. It doesn't ask the same basic questions over and over.

      Next Steps

      • Gmail integration: Sales Assistant sends approved emails automatically

      • CRM: emails for 340 clients. Most are still missing email addresses

      • biletti.fi collaboration kickoff

      Technical Implementation — How to Build Your Own

      If you want to do the same, here's the path. It doesn't require coding skills, just patience and about half a day. Works regardless of industry — barber, hairdresser, photographer, construction entrepreneur, consultant, shopkeeper.

      What You Need

      • OpenClaw — the framework everything runs on. Installed on your own machine or server.

      • Discord account — free. You create your own private server.

      • Anthropic API key — get it from console.anthropic.com. Deposit €10, lasts for several months.

      Step 1: Install OpenClaw

      npm install -g openclaw openclaw init

      This creates a workspace folder where your agent's memories and instructions are stored.

      Step 2: Create Discord Bots

      Go to Discord Developer Portal → New Application.

      For each agent:

      • Create a new Application (give it a name — e.g., "Assistant" or "Salesperson")

      • Bot tab: Reset Token → copy the token

      • OAuth2 → URL Generator → select bot + Administrator

      • Invite the bot to your own server with the generated link

      Step 3: Configure the Agents

      Each agent is a folder containing a SOUL.md — this is the agent's "personality and instructions". The most important thing is to write in it:

      • Who it is and what its task is

      • What it knows about your business (products, services, prices, customer base)

      • How it behaves — direct and concrete, not "maybe you could consider"

      Example SOUL.md:

      SOUL.md — Sales Assistant

      I help [your name] sell [your service/product].

      Work Style

      • I give concrete suggestions, not generic advice
      • I know the customer base: [who they are, what they need]
      • I draft outreach messages that [you] approve before sending

      Business

      [Write everything essential here: what you sell, to whom, at what price, what your competitive advantage is, how you contact customers]

      This is where you spend the most time — the more precisely you write, the more useful the agent becomes.

      Step 4: Configure OpenClaw

      Add agents and Discord tokens to the OpenClaw openclaw.json file:

      { "agents": { "list": [ { "id": "assistant", "name": "Assistant", "model": {"primary": "anthropic/claude-sonnet-4-6"}, "workspace": "/.openclaw/workspace-assistant" }, { "id": "sales", "name": "Sales", "model": {"primary": "anthropic/claude-sonnet-4-6"}, "workspace": "/.openclaw/workspace-sales" } ] }, "channels": { "discord": { "accounts": { "assistant": { "token": "DISCORD_BOT_TOKEN_HERE" }, "sales": { "token": "DISCORD_BOT_TOKEN_HERE" } } } } }

      Step 5: Automations (Crons)

      Want the agent to do something automatically — e.g., list daily tasks every morning or remind you about clients you haven't contacted? It's done with one command:

      openclaw cron add
      --name "MorningReminder"
      --cron "0 8 * * 1-5"
      --tz "Europe/Helsinki"
      --agent assistant
      --channel discord
      --message "Check if there are open tasks or clients that need contacting..."

      How Much Time Does It Take?

      Realistically:

      • Discord bots created and on the server: 30 min

      • OpenClaw installed and configured: 1-2h

      • SOUL.md files written (this is the most important step): 2-4h

      • Crons and automations: 30 min

      Total about half a day. The most important work is the SOUL.md files — the more time you spend writing the agent's context, the more useful it becomes. A generic agent is useless. An agent that knows your business model is valuable.

      Want Help?

      If this interests you but you don't know where to start — send me an email. I'm happy to help you get started.

      nabil@kivamedia.fi

      And if you want to know when I publish the next article, subscribe to the newsletter below. No spam, I promise.

      Nabil Samari is a Helsinki-based entrepreneur, DJ and event producer. Building automations and digital tools for small businesses — kivamedia.fi

      #Coding#AI#Entrepreneurship#Tutorial#Guide#Technical

      Subscribe to newsletter

      Get the latest posts straight to your inbox.

      No spam. Unsubscribe anytime.

      Latest
      I Built an AI Lead Pipeline That Isn't Allowed to Send Anything
      I Built an AI Lead Pipeline That Isn't Allowed to Send Anything
      August 10, 2026
      OpenAI Wants to Become Part of Everyday Life. Here Is How I Route My AI Work
      OpenAI Wants to Become Part of Everyday Life. Here Is How I Route My AI Work
      July 31, 2026
      I Tested Five Local Coding Models — Not One Built a Working App on the First Try
      I Tested Five Local Coding Models — Not One Built a Working App on the First Try
      July 29, 2026
      I Built a Local RAG Search. Having the Right File Wasn't Enough
      I Built a Local RAG Search. Having the Right File Wasn't Enough
      July 29, 2026
      Popular tags
      AIYrittäjyysHenkilökohtainenKoodausTapahtumatTekoälyTeknologiayrittajyysTutorialGuideTechnicalADHDProjektiTyönhakuFestivaalitPälkäneen KesäpäivätValokuvausYökerhotKalustoVinkitLuova alaLuovuusSEO
      Subscribe

      Get new posts straight to your inbox.