AI Agent Execution Beta
The AI Agent
That
Does It
for Your Users
With your user’s consent, the assistant carries out approved workflows itself — filling fields, clicking through steps, and narrating as it goes. It verifies the outcome before it says done. Off by default; you switch it on per workspace.
In beta · Off by default · Every run needs your user’s consent
How a run works
Consent, then action, then proof
Your user approves the plan
The assistant shows exactly what it’s about to do. Nothing runs until your user explicitly says go — every run, every time.
It acts, and says so
It fills fields and clicks through the approved steps, narrating each one as it goes. Your user watches it happen and can stop the run at any time.
It proves it worked
A run isn’t done when the clicks are done. The outcome is checked against what success looked like in your recording before the agent says done.
Trust first
Guardrails first, autonomy second
Acting in your product is a privilege the user grants each time — never a default. The guardrails aren’t settings to remember; they’re how a run works.
- Off by default — you accept the acting terms and enable it per workspace
- Only workflows you approved can ever run
- Every run starts with your user’s explicit consent
- Sensitive values are typed by the user into your own fields — never through FlowBuddy
- Narrated live, stoppable at any time, outcome-verified before "done"
The application intelligence layer
One layer powers all of it
This isn't a separate tool to set up. Every capability here reads the same application intelligence — the workflows, concepts, and screens FlowBuddy learned from one recording of your product, approved by you.
See how the layer is builtApplication Intelligence
Also powered by this layer
Give your users the help and support when they need it the most.
FAQ
Questions about agent execution
Is AI Agent Execution available today?
Yes, in beta. It is off by default: you accept the acting terms in Studio and switch it on per workspace. Until you do, your assistant stays read-only — answering and guiding, never acting.
How does consent work?
Per run, not per session. The assistant shows the user what it is about to do; the run starts only when the user explicitly approves it, and the user can stop it at any moment. Only workflows you approved can run at all.
What about passwords and sensitive fields?
The agent never touches them. Sensitive values are always typed by the user directly into your product’s own fields — they never pass through FlowBuddy.
How do I know it actually completed the task?
The run verifies its outcome before reporting done: it checks the page for the evidence of success your recording established — and if your product shows an error instead, the run says so rather than claiming success.
More questions? Read the full FAQ