Overview
About Automation Anywhere.
THE COMPANY
Automation Anywhere builds digital workers, software bots that automate repetitive office tasks across the world's largest enterprises. Over 400 million automations run on the platform every year.
THE PRODUCT
AI Agent Studio is the GenAI tooling layer inside Control Room, the developer-facing interface of the platform. It lets developers build AI Agents that perceive, reason, and act inside automation workflows.
Problem
Same mental model, new power.
Automation Anywhere runs 400M+ automations a year for the world's most compliance-heavy enterprises. Whatever we added couldn't break the trust the platform had already earned.
And the people building those automations had never written a prompt. They build from pre-built actions and templates, predictable pieces that run the same way every time. GenAI asked them to work in a way they'd never had to learn.
The goal was to let any developer build effective AI workflows without learning prompt engineering first.
Ethics
Security
Privacy
Reliability
Accessibility
Transparency
Accountability
Process
The approach.
Field study
Learning from our competitors.
Every major platform was shipping their version of agentic AI. We studied the field to understand what was working, and what wasn't.
Feature comparison matrix — 7 dimensions across 6 platforms
What we picked
01
Model selection
3-step process: Provider → Model → Version. Easier for non-technical users than a single dropdown of model+version combined.
02
Prompt input
Natural language builder with inline variables. Prompts read like prose, not code.
03
Versioning
Named versions with descriptions of what changed. Not date stamps, not auto-incremented numbers.
04
Cost visibility
Live token count and cost estimate per prompt. Visible at build time, before anything ships.
Personas
Different goals, one handoff.
The user personas and their roles across the end-to-end journey, from setting up to governing.
Roles and permissions, Automation Anywhere docs →
Minh
wants AI adopted, safely
Jake
wants models ready and tracked
Marcus
wants templates others reuse
Sue
wants to build without prompts
Rochelle
wants proof of every use
Lead
Minh
Asks for AI in automations, with the data watched.
"We're careful about what leaves with the models."
Admin
Jake
Creates the model connection and turns on the governance log.
"Get the models ready, and log every use."
Pro Dev
Marcus
Builds a reusable prompt template, publishes it to the shared repo.
"The team gets GenAI without writing prompts."
Citizen Dev
Sue
Pulls the template into her automation, new or existing.
"No model setup, no prompt engineering. I just use it."
Rochelle, GRC — audits prompts and model usage across the whole chain.
Reviews prompt and model usage across all of it.
"Prove what ran, and that nothing leaked."
User flows
How each persona uses it.
With the personas defined, we mapped their steps through the feature. Five flows, in the order they happen.
flow 01 · Govern
Set up governance.
Jake turns on AI data logs for the org. Everything AI does from here gets recorded.
click to zoom
flow 02 · Connect
Connect a model.
Jake connects the models everyone builds on, tests them, and decides who gets access.
click to zoom
flow 03 · Create
Create a prompt template.
Marcus builds a template in the workbench and publishes it for the team.
click to zoom
flow 04 · Use
Use it in an automation.
Sue picks the template in her automation, fills the variables, and runs it.
click to zoom
flow 05 · Monitor
Monitor every run.
When Sue's automation runs, the record writes itself. Prompt logs by session, event logs for every step.
click to zoom
How it works
One handoff, three systems.
Once they hand off, every request the agent makes gets built, checked, and recorded, by three separate systems.
swipe to explore →
The agent is one step inside an existing automation. Every request it makes runs through all three systems first.
The build surface
A glimpse of AI Agent Studio.
Tap a numbered dot to learn more
The audit trail
Understanding AI governance.
Marcus wants to know why an automation failed. Rochelle needs proof that no data leaked. One Session ID answers both.
What it covers
Audit logs give one centralised record of user activity and automation events. AI governance builds on that record.
- Usage insights across prompts and models
- Permission based access to audit logs
- Raw prompts, responses, variables and feedback, all visible
- Personalised AI governance dashboards
- Search and export to SIEM platforms
The logs
Before any screens, we mapped what deserves an event: every use case broken into its audit events, with encryption decisions and what shipped in phase one.
click to zoom
From templates to agents. The same surface now audits full AI Agent runs step by step, with tool calls, prompts, and token usage captured live, on the permission model this design set.
Impact
How it's going so far.
Model usage captured by internal tool
