If you haven't heard about Product Club — the Pathfinder Community Product Club is a monthly virtual meetup led by Automation Anywhere product leaders that focuses on our latest proprietary product innovations. It offers a place for community members to stay informed, connect with product leaders, and gain insights into real-world applications of the latest innovations in intelligent automation. Stay updated on WHEN Product Club’s will go live each month by subscribing to our events page.

P.S. If you can’t attend a meeting, no worries — we'll be dropping a recap of each month's session right here in our Product Club hub.

HOSTS

Oli Morris, Developer Evangelist
Nishikanth N, Director of Product Management
 

TOPIC

In this super special Product Club Super Session, we discussed how to build fail-proof automations at warp speed with the advanced capabilities of Automator AI.

Here’s a rundown of the session:

1. Nishikanth takes us through the products and feature sets encompassed by Automator AI.
2. Nishikanth performs live demos of Autopilot, Co-Pilot for Automators, and Generative Recorder.
3. Oli and Nishikanth answer live audience questions.

THE POWER OF AUTOMATOR AI

Automator AI spans the automation lifecycle to accelerate process discovery, development, and maintenance of automations, making the build process 30% faster and automations 50% more resilient.

Powered by Automation Anywhere’s natively trained LLMs from over 300 million automation runs, the model is completely grounded to our metadata.

The Automator AI suite is comprised of:

• Autopilot - identifies and understands every process. It helps maximize ROI by prioritizing the highest impact opportunities. It can take PDDs or BPMN files as inputs and transfers those into an automation outline, helping you reduce discovery time from months to days.
• Co-Pilot for Automators - allows you to input automation goals in natural language and converts these instructions into automations, as well as uses AI to suggest the next best set of actions.
• Generative Recorder - helps you build resilient, self-healing UI-based automations and reduces downtime by up to 50%. When a source application interface changes due to updates, generative AI identifies that change and updates the automation in real time.

AUTOPILOT & CO-PILOT FOR AUTOMATORS DEMO

An automation admin has uploaded a BPMN file for order management. Ordinarily, converting this end-to-end series of automations would take a few weeks.

1. Automation Admin uploads the BPMN file and assigns it to a developer to build the automation.
2. Developer has permission to use AI, so he clicks a feature called “Generate Automation.” After a few seconds, he clicks ”Refresh” and Autopilot has created an outline of multiple automations from the BPMN file with a process that encompasses all the steps at the top.
3. The developer opens the process and can see the outline which includes all task bots, conditions, and necessary APIs or sub-processors based on the information in the BPMN diagram.
4. The developer goes back and opens one of the task bots, “Confirm Salesforce Order.” He sees this task bot has been created with all the best practices in place including exception handling and logging. The developer can still drag and drop additional necessary actions and packages in.
5. He opens Co-Pilot for Automators to enhance the task bot. The Co-Pilot Assistant window opens and he is able to input instructions in natural language: “Connect to salesforce. Read orders from C:\InvalidateOrders.xlsx excel and update each record in salesforce.” The Assistant converts the instructions into appropriate actions and inserts them into the automation, and the developer can still review and make changes.

GENERATIVE RECORDER DEMO

In many cases, developers are creating automations that log into SaaS applications such as Salesforce, Servicenow, or Workday. As we know, SaaS applications will always undergo changes, which can mean failures in productions, breaches of SLAs, impact on the return of automations, etc. Let’s take a look at how Generative Recorder can identify the point of failure and help you recover to continue your automation.

We have an automation that logs into Salesforce. We’re going to observe 2 scenarios:
 

1. Generative Recorder is not enabled:

• If you click on Advanced Settings within the automation, scroll down to Package Settings > Recorder and expand the section. The option under Generative Recorder for Generative AI-based fallback is unchecked.
• Now we run the automation. It logs into Salesforce, carries out a few actions, and hits a roadblock when it doesn’t find an element we defined as part of the automation. We receive an error message and the automation terminates.

1. Generative Recorder is enabled:

• Return to Advanced Settings within the automation, scroll down to Package Settings > Recorder and expand the section. The option under Generative Recorder for and ensure the Generative AI-based fallback is checked.
• We run the same automation again. It logs into Salesforce, carries out a few actions, and hits the same roadblock. However, this time we receive a message with information on what the automation was not able to locate, plus the path we should look at that will fix the issue. Now, we are presented with a choice: 1) we can stop the automation and go back to fix it, or 2) let it auto-heal and continue running the automation.
• Because we are in developer mode, we see the model window with this information. However, in the case of production, Generative Recorder would auto-heal and continue with the automation, then send the admins or developers a notification on the failure point and how they could fix that failure.

SESSION Q&A

Thank you to our audience for submitting their questions! Unfortunately, we aren’t always able to answer them all during the live session. We want to express our gratitude to our special co-host, Nishikanth, for providing his expertise and responses.

**Please note that all answers were shared live during the January 2025 meeting, and are subject to change. We strongly encourage you to contact your account management team for any specific licensing and pricing inquiries.

Q: Is this feature (Generative Recorder) available for on prime deployments?
A: Not yet, at the moment it is cloud only.

Q: What is needed in order to use the AI fallback?
A: We've got a handy page in our documentation here which breaks down all of the components in automation anywhere and says whether there's a license required, whether they need the clouds and, or whether they're on premise as well.

Q: Does the generative AI model continuously learn from failed automation attempts in production? And if so, how does it ensure that this learning doesn’t introduce new biases and errors?
A: That’s a fantastic questions. So here’s what happens: we give a configuration in generative AI where you can decide if you really want the generative AI to take the corrective action by itself or you would just want it to identify and notify you so that you can take a look at it, do a due diligence, and then fix it. So even if it is continuously learning, we leave the choice in the hands of the developers or the admins to decide how they want to go about it, because we do know not every customer, and not every developer is going to have 100% faith in the generative AI models. That’s why we give flexibility to developers and admins to make a call whether they would want to look at the suggestions and then make a fix in the automations.

Q: If I recall correctly, the generative AI recorder was made to work with Azure Open AI. Anything on the road map for you said like native Google support, but equally expand that question say for any other LLM support?
A: Yes. We are keeping the options open. In fact, you may have noticed Claude has recently released vision models, so we are in the early stages of evaluating those. So, we have open eyes on all the models coming out in the market including Google as you mentioned.

Q: Does Generative Recorder make the correction only on the Xpath property of the object or can it predict the failures and correct other properties like ID and class?
A: Apart from X path, let's say if you change the label of login ID to username, it'll identify those kind of changes as well. So yes, it in fact tries to find out what the similar labels or fields are in the screen and tries to match to that. That is why I said it's always also a good choice to see the recommendation and then make the changes. It takes multiple things into consideration when it gives the suggestions.