Running a US marketing agency
from a village in Spain.
The AI handles most of the work.

The Problem

I needed to run a US marketing agency from a village in Spain, without hiring a team. Every function of the business — copywriting, research, operations, sales coaching, coding, data analysis — needed to happen, reliably, every day. Human headcount for that is $8–15k/month and months of hiring. I had neither.

I also needed to see the whole thing working. AI agents that run overnight are worthless if I can't tell what they did, what they decided, or what needs my sign-off in the morning.

The Solution

Most people use AI the same way: open a chat window, type a question, get an answer, close the tab. That's not a team. That's a calculator.

I built a team. 10+ specialised agents — each with their own name, role, memory, and communication channel. Not tools. Not assistants. Teammates. Jess (the orchestrator) has her own Telegram group with me, her own workspace, her own daily memory logs. Alex (Sales Coach) has his own Telegram and Slack bots — John and Ashley, my sales reps, DM Alex the same way they'd message a human coach. Objection roleplay, script review, call prep. Mara (Operations) owns client onboarding and delivery integrity. The team messages her directly with ops questions. She remembers every client, every commitment.

The key architectural detail: each agent has an isolated workspace — its own files, its own memory, its own identity (SOUL.md, IDENTITY.md). Alex doesn't know what Mara is working on. That's intentional. Isolated context means no contamination, no confusion, no context bleed. And no starting from scratch every session — these agents remember what was decided last week.

I didn't build a chatbot. I built a team.

Mission Control is the visible layer on top: a real-time web dashboard that surfaces everything the agents produce — who did what, every task completed, every daily log, every doc, every decision that needs approval, every R&D memo. Six tabs. Readable on a phone in 30 seconds. The whole business, at a glance, from anywhere.

What each Mission Control tab shows:

• Agents — Every agent (Jess, Alex, Mara, lrf-coding, lrf-copy, lrf-research, etc.), their model, current task, readiness badge, and which channels they're active on
• Tasks — Every task completed across the business: agent attribution, status, dates
• Daily Log — Every day's memory file. What agents decided, built, learned. Complete audit trail
• Docs — Every skill, campaign template, case study, memo — searchable, Redis-backed
• Approvals — Anything agents want to do that needs human sign-off queues here. I review from my phone
• R&D Memos — A committee of 5 AI models debates strategy Mon/Wed/Fri at 9am, synthesises a memo, auto-publishes here

Tech Stack

The agent team runs on the OpenClaw runtime, with specialised roles powered by Claude (Opus, Sonnet) and GPT. Each full agent has a dedicated Telegram bot and Slack integration. Memory persists via flat files synced to Redis — personas defined in SOUL.md identity documents, not just prompts. Mission Control itself is Next.js 15 deployed on Vercel, backed by Redis (Upstash), styled with Tailwind. A Bash sync script on a cron job pushes workspace data to Redis every 2 hours so the dashboard stays live.

Timeline

Under 14 days. From zero to a full working AI agent team with its own dashboard, Telegram and Slack bots, daily briefings, and shared memory.

I didn't write the code. I described what I needed to my AI orchestrator, who dispatched each build to a Claude Code agent with full file permissions. I reviewed, gave feedback, merged. Incremental additions daily since.

Results

• 1 person (me) runs a US marketing agency from rural Spain with no offshore VAs, no full-time staff beyond 2 remote sales reps (John, Ashley)
• ~$500–$900/month total AI cost running the full team of 10+ specialised agents — vs. $8–15k/month for equivalent human headcount
• 3 full agents + 8 sub-agents across Telegram and Slack — John and Ashley DM Alex (Sales Coach) and Mara (Ops) the same way they'd message a human teammate. No prompt engineering required.
• 100% of business context (clients, campaigns, skills, daily logs, R&D memos, tasks) visible in Mission Control — I can see my whole business at a glance from my phone
• Daily 8am Telegram brief auto-delivered every morning — news summary, YouTube picks, my tasks for the day, what the AI orchestrator is running

The Problem

Most marketing agencies are built on people. SDRs sending emails. VAs logging data. Ops managers chasing follow-ups. That model breaks the moment you try to scale — and the cost compounds fast.

I needed to run a cold email agency for US loan officers at real scale: thousands of outbound touches per day, inbound replies handled instantly, leads qualified and booked without a human in the middle. Hiring the team to do that manually would run $33–52k/month. That's before you factor in management overhead, sick days, and turnover.

The other problem: most "agency stacks" are just a CRM and a Zapier account duct-taped together. One broken zap and the whole pipeline goes dark — usually at 2am on a Friday.

The Solution

I built the whole revenue engine from the ground up. Not just a dashboard on top of existing tools — the actual infrastructure that drives outreach, handles conversations, syncs data, and keeps the deal pipeline moving with minimal human involvement.

One lead enters the top. A qualified loan officer call comes out the bottom. Everything in between is automated.

What the stack handles end-to-end:

• Outreach at scale — Smartlead runs the cold email sequences. AeroSend manages 60–70 mailboxes across 20–40 domains, so thousands of emails go out daily without hitting spam filters
• AI conversation handling — Inbound replies don't sit in an inbox. AI agents parse intent, warm the lead, answer questions, and book the meeting. No SDR required
• CRM as single source of truth — Close CRM holds every lead, every call, every SMS, every task. Nothing lives in someone's head or a spreadsheet
• Automation backbone — Make.com orchestrates the data flow: reply comes in → AI parses it → Close updates → Twilio SMS fires → Airtable logs it → Slack notifies the team. Every step, every time
• SMS follow-up — Twilio handles text outreach to leads, triggered and managed by AI. Responses feed back into the same loop
• Data layer — Airtable tracks everything: 42 tables, 297+ fields on Clients alone. Campaigns, responses, SMS logs, closings, bookings, realtor service data — all of it queryable
• Integration glue — Calendly for booking, Loom for sales calls, 1Password for secure credential management across the whole team

Tech Stack

Smartlead, AeroSend, Close CRM, Make.com, Twilio, Airtable, Slack, Calendly, Loom, 1Password. AI agents (Claude) sitting in the conversation layer, parsing replies and driving next actions.

Timeline

Built incrementally over Q1 2026. Core infrastructure — outreach, CRM, automation backbone — was live within the first few weeks. The AI conversation layer and full data architecture came next. It's not a one-day build, but it's also not a 12-month IT project. You scope it right, you build it in layers, and it compounds.

Results

• 40,000+ US loan officer leads + tens of thousands of realtor leads in pipeline, being touched daily by a stack that costs $1.5–3k/mo to run
• Replaces $33–52k/mo in human salaries — 2–3 SDRs, a data analyst, an ops manager. The infrastructure does their jobs, without the overhead
• Thousands of cold touches + AI replies + CRM syncs per day with effectively zero human involvement between outreach and a booked call
• $999/mo offer — the whole agency runs on a lean price point because the cost of delivery is infrastructure, not people
• One operator (me) runs the full outreach-to-close pipeline from rural Spain, with two sales reps handling calls — and the stack keeps running whether I'm at my desk or not