# Can my AI assistant run load tests or start monitors?

> An AI assistant can write and play scripts, build load test scenarios, and set up monitors in Loadster, but launching a full load test stays your call.

Source: https://loadster.com/faqs/ai-assistant-run-load-tests/

Almost! Once you connect an assistant to Loadster over the
[MCP server](https://loadster.com/faqs/connect-to-loadster-mcp-server/), it can take care of the scripting and building the test scenario,
but actually launching the test or starting a monitor requires a button click from you.

This is because running load tests and monitors burns Loadster Fuel and generally costs money, so it's good to have a
human in the loop to prevent unexpected costs.

## What Your AI Assistant Can Do

Working through Loadster's MCP tools, your assistant can:

- **Write and edit scripts** for both [Protocol Bots](https://loadster.com/manual/protocol-scripts/) and real browser testing, from a
  plain-language description of the flow you want to test.
- **Play a single bot** to check that a script works end to end, and read back what happened to fix what breaks.
- **Build and save [load test scenarios](https://loadster.com/manual/load-test-scenarios/)** with bot groups, ramp patterns, and regions.
- **Manage [datasets](https://loadster.com/manual/dynamic-datasets/) and [monitors](https://loadster.com/manual/monitoring/)**, including configuring a disabled
  monitor and its schedule. You enable the monitor in the dashboard when you're ready for it to run.

Because the assistant works in your real Loadster project, everything it creates is a normal Loadster script, scenario,
or monitor that you can open and modify in the dashboard.

## Reviewing What the Assistant Builds

Scripting is usually the slowest part of load testing, so letting an assistant draft scripts and scenarios can save you
a lot of time.

Still, it's a good idea to review a generated script before you rely on it: check that the steps match the flow you had
in mind, that assertions catch the failures you care about, and that any
[wait times](https://loadster.com/faqs/think-time-wait-time/) are realistic.

Keep experimental scripts in their own separate Loadster project, and only point scripts at systems you're authorized to
test.