Why an agent needs managing at all
An agent is tested against the work it sees at launch. Six months on, the exceptions are different, a supplier has changed its email format, and the provider has replaced the model it runs on. Nothing is broken, exactly, but the answers drift. Managing an agent is the work that keeps it trustworthy: watching it, tracking what it costs, retesting it when something underneath changes, and tuning it as the work changes.
Running it yourself
You own the agent, the code, the cloud accounts and the documentation, so you can. The runbooks cover the routine things, and the alerts tell you when something stops. What your people need is someone to read the exception queue and update the rules when the business changes.
Ongoing support for one agent
Usually $250 to $3,000 a month, optional, and most projects sit at the lower end. It covers monitoring and alerts, fixes, hosting help, retesting when a provider changes its models, tuning as the work changes (including fine-tuning a model where that earns its place) and a set amount of improvement. You can end it on 90 days’ notice and keep the agent.
A fractional AI Automation Director for the whole program
When you run several agents and AI tools, someone has to own the whole picture. The retainer covers every agent and tool you have, including ones other vendors built: vendor and tool analysis, contracts, the roadmap and the AI use policy, spend, maintenance of completed projects and a monthly report. Support for the agents we built is included, so it isn’t an extra cost on top.
Related questions
- Who owns the code?
- You do, on final payment, in your own repository and cloud account.
- What if our developer disappears?
- The documentation, runbooks and your own accounts mean another developer can pick it up.