AI Agents, Explained by
Someone Who Runs Them
Not theory. Not hype. Just what works — from someone who runs 15 agents across 4 businesses.
What is an AI agent?
An AI agent is software that takes actions — not just answers questions. When you ask ChatGPT "what should I do about X," that's a chatbot. When software reads your financial data, reconciles it against your bank feed, flags discrepancies, and drafts the journal entries — without you touching it — that's an agent. The difference is execution. A chatbot gives you a to-do list. An agent does the list.
Think of it as a junior analyst who never sleeps, works in parallel with other analysts, and follows instructions exactly. Except this analyst can also use tools — call APIs, query databases, read and write files, send emails. The instructions you give it are called a prompt, and getting those instructions right is the highest-leverage skill in this space.
What is multi-agent orchestration?
One agent is useful. Multiple agents working together is a business. Multi-agent orchestration means coordinating several agents to tackle different parts of a problem simultaneously — one handling financial reconciliation, another managing social media, a third running competitive research, a fourth writing code. None of them touch each other's work. All of them finish.
The key insight: coordination beats capability. A single brilliant agent can't do four things at once. Four focused agents, each with clear boundaries, can. I call this an agent swarm, and it's the core concept behind everything in this newsletter. It's also the subject of Field Note #001.
Why should an operator care?
Cost asymmetry. A task that costs $150/hour with a contractor costs roughly $0.50/hour with an agent. Not for everything — agents aren't replacing your accountant or your lawyer — but for the structured, repeatable work that eats 60% of an operator's week? The math is brutal in your favor. And agents work at 2 AM. They don't need onboarding. They don't call in sick.
Speed is the other lever. Last month, agents handled our monthly financial reconciliation, social media calendar, SEO audit, and competitive intelligence brief — simultaneously. Four tasks that would take a human team a week, done in parallel in under an hour. If you're running a business between $500K and $10M, this is the gap where agents create the most leverage: you can't afford a team of 10, but you need the output of one.
The Stack
Every agent system has four layers. Here's the map:
Start reading
Field Note #001 is the story of building the system described above. Four agents writing code in parallel across four terminal sessions. What worked, what broke, and the pattern underneath.
Read: I Built an Agent Swarm That Builds Itself →Get field notes from the frontier.
Two essays a month. AI-first strategy, real numbers, tools you can ship today.