Building my first agent
The word "agent" gets used loosely. For our purposes here: an agent is a routine that runs without you, on a schedule or a trigger, producing an output you'd otherwise sit down and produce yourself. Not a chatbot. Not a one-shot prompt. A small persistent worker that does a job.
Most CEOs who want to build one stall in the same spot: they don't have the substrate yet. They're trying to skip from "I asked ChatGPT a question" to "I have an autonomous worker," and the gap between those two things is bigger than it looks. This path closes that gap in five steps. By the end you'll have shipped one real agent (the commitment ledger), seen how it's structured, and have the template for the second one.
About eight hours. Most of it is the install + first-run of the two workflows that anchor the path.
What you'll have when you're done
- A working mental model of what "agent" means in operator-grade AI versus the marketing version.
- Persistent memory across Claude Code sessions · the substrate every agent runs on.
- Your first autonomous routine running on a schedule, producing output without you.
- A commitment-tracking agent watching every promise made by you and to you.
- A pattern you can copy for the next agent you want to ship.
Unlock the path
Members get the connective prose for every step · why it's here, what to look for, what you'll have when you finish. Membership is becoming one All-Access plan: every book, every course, the templates library, and the Deep Dive archive. It opens soon, and newsletter readers hear first.
The path
-
1Claude CodeRead it →
Start here for the substrate. Claude Code is the CLI tool the agents on this site run inside · not the consumer Claude.ai chat. The distinction matters because agents need a runtime that can hold state, run on a schedule, and read your local files. Read the definition. Note the difference from the chat app you've used before.
After this you'll know what kind of tool you're actually building inside of.
-
2Persistent memory across Claude Code sessionsRead it →
Agents need memory. Without it, every run starts cold and every output looks like a one-shot prompt. With it, the agent gets sharper every week: it remembers the team, the projects, the recurring patterns. This workflow installs persistent memory across Claude Code sessions · the substrate every later step needs.
After this you'll have a Claude install that holds state across runs, which is the first requirement for anything autonomous.
-
3Run autonomous workflows 24/7 with Claude Code RoutinesRead it →
This is where "agent" starts to mean something concrete. Routines are autonomous workflows that run on a schedule · the agent shape, sized down to one specific job. The workflow walks you through setting up the first one. Don't aim for ambition here. Aim for "ships and runs by tomorrow morning." A small agent that actually runs beats a perfect agent that doesn't.
After this you'll have one routine wired up and running on its own.
-
4The commitment ledger: every promise you made (or made to you) trackedRead it →
Now the first real production agent: the commitment ledger. It listens to every meeting, logs every commitment made by you and to you, and follows up automatically when something goes overdue. This workflow is the most concrete example on the site of what a CEO-grade agent looks like in production. Follow it carefully · the structure (input source · state · trigger · output) is the template for the next agent you build.
After this you'll have a commitment-tracking agent in production AND the pattern you'll use again.
-
5Ask your meeting history anything: the CEO's query layerRead it →
The second agent. The query layer over your meeting history is a different shape from the commitment ledger · less "react to events," more "answer questions on demand" · but it shares the same substrate (memory, routines, transcripts). Build this one to internalize that you now have a kit, not a one-off.
After this you'll have shipped two agents · which is when the third one stops feeling hard.
What's next
When this path is closed, the natural next commitment is 30 Days to Real AI Leverage (Course #1) for the deeper systematic walk. If you want a sibling path before the course, try I have ChatGPT, I want real leverage for the workflow-focused install or I'm new to AI · 10 hours that change everything for the conceptual ground.