# Maestro Cue: Agents That Pick Up Work on Their Own If you are looking to make a real leap in AI adoption maturity, the ladder goes something like this. First you learn to chat with an agent. Then you learn to hand it a whole task and let it run. Then you graduate to spec-driven developer (Auto Run Playbooks in Maestro), where you write the steps down once and the agent executes them on demand. The next rung is the one most people have not climbed yet: how do you get the AI to pick up work without you telling it to at all? That is the question Maestro Cue answers. Cue is a trigger-based agent pipeline engine ([docs here](https://docs.runmaestro.ai/maestro-cue)). You define what should happen and what should set it off, and your agents start acting on their own: on a schedule, when a file changes, when a pull request lands, when another agent finishes its work. You can build these pipelines three ways, and all three edit the same thing: a visual editor where you drag and connect nodes, a YAML file if you like config, or, the way most of Pedram's get built, by just telling an agent in chat what you want and letting it wire the pipeline for you. Here is the part worth slowing down for. The agent inside Maestro is a complete power user of Maestro. It knows Cue end to end. It knows Auto Run. It knows the settings, the layout, the theme picker, every control in the app. So when you describe an automation, you are not asking it to write config and hope. You are asking a fluent operator of the tool to go do the thing you would have done by hand, except it does it in seconds and does not get the schedule format wrong. The best way to show what that buys you is to show you something real that runs on it. ## The Marketing Department Nobody Coded Maestro markets itself. Not entirely, but the operation that keeps its presence alive on X runs on Cue, built by an agent, working on its own. It started as a conversation. The ask was roughly: every morning, find what people are talking about in our space, have a few agents argue out the best angle for the day, draft posts in our voice, queue them to publish, then look back and measure what landed and keep the dashboard current. Nobody opened the YAML. Nobody dragged a node. An agent that already knew how to drive every control in the product went and built the whole thing: picked the triggers, set the schedules, named the stages, connected the pieces, saved it. What came out the other end is a running marketing department. Here is what it looks like: ![[2026-XX-XX-maestro-cue-1781979468723.png]] Every card is a job that fires on its own. They run on schedules, early in the morning and through the day, whether or not anyone is logged in. Read it left to right and you can follow the entire operation without knowing a single thing about the format it is stored in. It reads like an org chart because that is what it is. - **Discover** wakes up in the morning and figures out what is worth talking about: trends, conversations, threads that are heating up. - **Campaign Council** is a deliberation step. Several perspectives weigh in on what the day's campaign should actually be, instead of one agent guessing. - **Draft** turns the decision into actual posts, in the right voice, ready to ship. - **Post** publishes. Notice the two cards that converge into one publish step: a fresh draft and a carried-over one land together, so nothing in the queue gets dropped on a quiet day. - **Measure** looks back at what went out and scores what landed. - **Enrich Dossiers** keeps the files on key people and accounts warm, so the next campaign starts from context instead of a blank page. - **Refresh Dashboard** regenerates the reporting so there is always a current view to look at. - A **watchdog** and a **heartbeat** keep the whole thing alive and flag anything that stalls. Discover, decide, draft, publish, measure, repeat. A marketing team's week, compressed into cards that run themselves. None of it needs anyone awake. You look at the dashboard when you want to, not because the machine is waiting on you. And when you want to change it, you do not open the YAML and hunt for the right field. You tell the agent "add a step that replies to mentions" and watch a new node appear on the canvas. ## Anything You Can Click, an Agent Can Do Cue is one example. It is not the boundary. Anything you can accomplish in Maestro by pointing and clicking, an agent can accomplish by being asked. Spin up a new agent and point it at a project. Write an Auto Run playbook and kick it off. Reorganize your workspace. Change the theme. Adjust the settings. The agent is not bolted onto the side of the app reaching in through a keyhole. It lives inside the app and has the run of it. That is what makes the marketing pipeline above unremarkable to build and remarkable to look at. There was no special integration, no plugin, no glue code. Somebody described a marketing department and an agent that already knew how to drive every control in the product went and built one. ## What You Would Build Forget marketing for a second. Marketing is just the example that happens to be running. The same move works anywhere you would otherwise reach for a pile of scripts and cron jobs: - **A research desk** that scans your sources every morning, has a council of agents argue out what actually matters, and leaves you a briefing before coffee. - **A release pipeline** where a pull request showing up triggers review, a test pass, and a draft of the changelog, all without you asking each time. - **A personal CRM** that keeps itself warm: notices who you have not talked to in too long, drafts the outreach, and reminds you. - **A content engine** that turns one long piece into a week of posts, schedules them, and tells you which ones worked. - **A watchdog** for anything you care about that pings you only when something is actually off. You do not start any of these by learning a configuration format. You start by describing what you want to a thing that already knows how to build it. For a second worked example, [the Maestro message bus](https://pedsidian.pedramamini.com/Claude/Blog/2026-05-28-maestro-message-bus) is a Cue pipeline that puts an agent on the other end of an iMessage thread, deciding what to answer and when. ## The Shift The old way to automate was to learn a tool, write the wiring yourself, and maintain a graveyard of half-broken scripts. The Maestro way is to describe the outcome and let an agent that lives in the app assemble the machinery. The marketing department above is what that looks like when you let it run for a while: a real operation, working on its own, that began as a few sentences. You are not configuring software anymore. You are delegating to something that already knows how the software works. Tell it what you want the machine to do. It will go build the machine. Want to kick the tires? Grab [Maestro](https://maestro.sh) and read up on [Cue](https://docs.runmaestro.ai/maestro-cue). — Pedsidian #claude