Your app makes an LLM call and gets a paragraph back. It could start a real agent instead and get finished work back.
I will show the pattern I am building on now: the app spins up an agent harness in the cloud, hands it a job and a workspace, and collects the result. Files written, a repo updated, a research pass completed, a browser driven to the end of a task. The same harnesses you already run on your machine, running per request, per customer, inside a product.
What we cover in the hour:
- The service layer. One endpoint in your app that starts a cloud harness, passes it a skill and a workspace, and streams back what it did.
- Writing the job. How to specify work so a harness finishes it unattended, and where to put the stop so a person still approves anything that sends or spends.
- Cost and pricing. What a run actually costs, when a harness beats a plain model call, and how to price per job or per seat without guessing.
- The offer. The same rig pointed at a client workload and sold as a service, which is a different business than selling seats.
Bring an app or an idea where a single model call is doing work it is too small for. We will spend the back half on your builds.
Members only, on Zoom. Replay is posted in the club afterwards.