In case 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.
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
Allison Able, Pathfinder Community Director
Pratyush Garikapati, Director of Product Management
Nishikanth N, Director of Product Management
TOPIC:
February’s product focus was Automation Co-Pilot for Automators - an on-demand intelligent automation assistant for automation developers.
Here’s a rundown of the session:
- Pratyush reviews the power of Automation Co-Pilot for Automators
- Nishikanth walks us through a demo using Process Composer and Automation Co-Pilot for Automators
- Nishikanth reviews the major product capabilities of Automation Co-Pilot for Automators and Suggest Next Actions
- Pratyush and Nishikanth answer live audience questions
If you’d like to watch the entire session recording with visuals from the demos, you can view it here.
Want some additional context on Automation Co-Pilot for Automators? Check out the Getting Started Guide here.
PATHFINDER ANNOUNCEMENTS!
- NEW Automation Anywhere University courses are LIVE!
- You can now access one-time sign-ups for Product Club here
- Don’t miss our upcoming Developer Meetups & User Group sessions - RSVP here
AUTOMATOR AI FOR CLOUD
To review the Automation Success Platform: this is your existing A360 capability, where you discover, build, and scale your automations. What we are doing now is also making it your system of work for generative AI in 3 ways:
- Embedding custom-trained and out-of-the-box products in our models
- We are making a platform more open and extensible so you can leverage any public or private large language model you choose.
- We are building in guardrails to prevent misuse, ensuring that as you interact with AI models, you have the proper guardrails in place as you scale the capability across the organization.
To that end, we have 3 cutting-edge products available to you: Automation Co-Pilot for Business Users, Document Automation, and, part of the focus of today’s discussion, Automator AI.
Automator AI is designed to accelerate developers across the automation lifecycle.
A typical automation lifecycle comprises 4 key stages: discovery and definition, development, testing and acceptance, and production support and change management. We have created an entire automation lifecycle powered by generative AI so that CoEs and business teams can:
- Accelerate discovery with Autopilot
- Reduce automation building time with Co-Pilot for Automators and Suggest Next Actions
- Decrease time spent on documentation
- Improve SLA adherence by cutting down on automation failures with Generative Recorder
These capabilities make up what we call Automator AI for Cloud. Automator AI is our vision to provide critical products and capabilities developers require to build automations and accelerate their automation program faster and better. This includes Co-Pilot for Automators, which enables professional and citizen developers to build automations by simply describing them in natural language.
Moving further down in our roadmap, we have plans to launch Test and Documentation Generation to assist in testing, explaining, and documenting your automation, as well as Generative Recorder, designed to give your automations the ability to recover by adapting to application changes automatically to keep running and alert you to needed updates.
If you have questions about our different A360 capabilities, their licensing requirements, and availability on the cloud vs on-prem, we encourage you to reference our new helpful guide here.
DEMO: ORDER MANAGEMENT
A developer, Marcus, receives a change request to notify the Inventory Management Team. He also extends the automation to check for entitlements before approving an order. Let's walk through how he accomplishes these changes with Process Composer and Automation Co-Pilot for Automators:
- Marcus has the Order Management process already built out in Process Composer. To address the change request, he will start by clicking on Co-Pilot at the top of the screen. The Co-Pilot Assistant box will open, and Marcus will input instructions into the message box. So per the change request, he instructs, "Add a step for the Send Catalog process. Then email the Inventory Management Team using Send Notification bot."
- The Co-Pilot Assistant perfectly adds the actions that send the catalog and send the notification to the process.
- Marcus can continue to make changes as needed by dragging and dropping elements from the left panel or by giving more instructions in the form of natural language.
- Then, Marcus needs to make changes to the Check Entitlement automation. So, he opens the automation, and we can see he has already built out a few things where it authenticates the Salesforce connection and gets records from Salesforce. Also, to ensure that the content he's getting out of this entitlement is readable, he's using generative AI, in this case, Vertex AI, to consolidate the content from Salesforce and return it in a format that users can read.
- Once again, he opens Co-Pilot and types instructions "if promptresponse is 'Allowed,' then first connect to database server ItemQtyLimit with admin/password credential followed by database read and then disconnect database. If not, first open excel C:\RestrictedItemOrders.xlsx, find the next empty cell, and set it to OrderID." This is a longer instruction, so it may take a minute to process.
- Based on the instructions, Co-Pilot automatically adds the conditions to the automation. Marcus can continue to build on top of these actions if he wishes.
- Marcus feels that there may be additional actions needed to close the loop with respect to the operations of Excel. To help him do that, he can use the generative AI-based capability, Suggest Next Actions, which looks at the entire context of the automation and, based on the context, suggests the next set of actions that logically make sense. In this case, it suggests "Excel advanced: Close" and a few other actions that Marcus is not interested in adding. He unchecks those specific actions, then clicks Add to apply those desired actions to the automation.
The best part about this is being able to give instructions in the form of natural language. Whenever the instructions are given, it ensures that the output is always related to automation. If you provide any instructions that are not related to automation, it gracefully rejects them.
Additionally, this Co-Pilot is a non-inclusive assistant—meaning you can use it only when you want. You do not need to use it exclusively to create processes.
PRODUCT CAPABILITIES REVIEW
Co-Pilot for Automators:
- Create or edit an automation in Process Composer
- Map dependencies such as Process, Bots, or Forms in Process Composer
- Create or edit an automation in Bot Editor
- Update attributes of an action in Bot Editor
- Automatically create variables for mandatory attributes of an action in the Bot Editor
- Reject instructions not related to an automation
- Respond with user-friendly messages
- Accept feedback on the output it produces
Suggest Next Actions:
- An option to view suggestions on the next possible actions, based on the business context, in the Bot Editor
- Guardrails in place to restrict actions based on the package RBAC or if the packages are not available in the control room
- Flexibility to selectively add the actions
- Option to regenerate different sets of suggestions
For full A360 capabilities, licensing requirements, and availability on the cloud vs on-prem, please reference our new helpful guide here.
POLLS & RESULTS
- We asked: What is Automation Co-Pilot for Automators?
- Almost every audience member answered “an intelligent assistant to build automations using natural language,” which is correct! Automation Co-Pilot for Automators is not a chat widget to collaborate with other developers (though you can do that in our forums) or a search widget for product documentation (which you can find here).
- We asked: What are the primary challenges you often face in the development of automations?
- The top challenges for our audience were a tie between “lack of best practices/standards or their inadequate incorporation” and “frequent need for automation maintenance due to changes in underlying applications.” A great number of people also answered “lengthy time required for the automation build process,” “maintenance difficulties stemming from outdated documentation misaligned with the current state of automation,” and “time-consuming training of developers.”
- We asked: Do you see value in using Automation Co-Pilot for Automators within your organization?
- Almost every audience member answered yes! 50% believe it would be most valuable for their developers and 43% believe it would be most valuable for citizen developers.
- We asked: What enhancements would you like to see in Automation Co-Pilot for Automators?
- The audience was 50/50 split between generative AI-based debugging capabilities and the ability to generate documentation for automations.
- Finally, we opened up the poll for the live audience to enter their suggestions for crucial features, functionalities, and/or improvements that should be added to the Automation Co-Pilot for Automators roadmap. Here’s what they said:
- Testing and best practice suggestions for existing code
- Debugging, so when an error generates during a bot run it will return a natural language description of what the problem could be
- To make it available on-prem sooner
- Suggest and generate error handling and standard redundancy processess
- To adjust quickly to sudden unexpected changed in vendors site
- To interactively work with the reorder function to be able to identify UX/UI changes and be able to adapt and find the button on the screen that is used as an anchor
SESSION Q&A
Thank you to our audience for submitting their questions! Unfortunately, we couldn't answer them all during the live session. We also want to express our gratitude to our product leader co-hosts, Pratyush Garikapati and Nishikanth N, for providing their responses.
**Please note that all answers were shared during the week of February, 2024, and are subject to change. We strongly encourage you to contact your account management team for any licensing and pricing inquiries.
Q: Why is it named Co-Pilot and are there any limitations?
A: It’s called Co-Pilot because it is not meant to replace a developer. At the end of the day, the developer knows exactly what the automation is being built for, whether it meets business requirements, whether it’s working fine, etc. Our goal with Co-Pilot and broader Automator AI capabilities is to be able to accelerate a developer’s work. And that's been the consistent theme you will hear from us—making sure that developer has the right tools and capabilities to accelerate across the automation life cycle. The demo we showed today may not necessarily be days or weeks worth of change, but it should give you the sense of how you can use copilot to build out those capabilities.
As far as limitations and accuracy, one thing you may have noticed today are the prompting techniques that we have right now require you to be slightly more nuanced. Whether it's a change or building a new automation, you do need to be a little more nuanced around the key steps you'd want Co-pilot to add in your automation. Over the course of the roadmap, we plan to make this more abstract so that you don't have to provide granular instructions. That means you can potentially go and say, “build an order management automation,” and it will have a clear idea of how to do that.
Q: What's the accuracy level of Co-Pilot?
A: It's ultimately the developer who's gauging whether the automation meets the business requirements. Your accuracy has to be compared against whether the automation is doing what the business actually wanted, what was prescribed in the PDD or SDD documentation. That's essentially where the accuracy comparison comes from. Our intent to make meaningful improvements in Co-Pilot for Automators and Suggest Next Actions to improve its ability to understand the prompt or context of the automation and give you an output in the form of an automation or a suggestion.
Q: What's the accuracy level of Co-Pilot?
A: It's ultimately the developer who's gauging whether the automation meets the business requirements. Your accuracy has to be compared against whether the automation is doing what the business actually wanted, what was prescribed in the PDD or SDD documentation. That's essentially where the accuracy comparison comes from. Our intent to make meaningful improvements in Co-Pilot for Automators and Suggest Next Actions to improve its ability to understand the prompt or context of the automation and give you an output in the form of an automation or a suggestion.
Q: Why is Automation Co-Pilot very primitive in terms of UI?
A: We would love to engage with you to unpack that comment a little more. What kind of capabilities would you want to see in the UI? If you can answer the poll, that will be helpful, or if we can engage with you, please drop in a line, or we would be happy to engage with you to get feedback and see what we can improve.
Q: If we have our own instance of an LLM model in your listing, do we need any special training or specific instance so it can build AAI code as you've demoed?
A: No. In the product strategy that I spoke about earlier when we showed the Automation Success Platform, these broader Automator AI suite of capabilities are where we are infusing large language models - both external and internal models - into the product, which is why it's not a bring-your-own-LLM scenario. It's multiple models at play and we have our own custom-tuned models that we have built-in and infused within the product. They run in tandem with external large language models which are currently SaaS-based, which is why Automator AI as a whole is a cloud-only product. This innovation is largely gonna happen first in cloud, and on-prem compatibility will come up later. That's only a matter of time.
Q: Will you be linking Automation AI to process discovery?
A: Yes, that's what Autopilot is all about, which is to discover your processes and then be able to export that PDD and the supporting documentation or supporting files into Control Room and use the Autopilot capability to convert the discovered process into an automation.
*FYI: Our next product club meeting happens to be on Autopilot! That listing is live on our community events page to RSVP.
Q: Can you help us understand how it works to get the text in the code in our Automation 360 tool?
A: I would direct everyone to our Security/Architecture page. It explains the data flow of how we take in a prompt and produce an automation or suggestions on the back of it.
Q: Is it possible to revert the additions in code and process done by co-pilot?
A: Yes, of course. Co-Pilot for Automators is an assistant to developer; the developer can always undo/redo the changes in code and process as desired.
Q: Does co-pilot use prompts and feedback from customer to retrain LLM? Does it get stored anywhere on AAI side?
A: We never use customer data for model training. Please refer to the Security/Architecture page in product documentation for details on how we handle prompts and feedback.
Q: We have latest A360 on-prem Setup - Automation 360 build 21023, but we are unable to see the Co-Pilot Feature available. Are there any restrictions for on-prem set-up?
A: Automator AI, including Automation Co-Pilot for Automators, is currently available only on the cloud and requires additional licensing. On-prem support is planned for CY 2025.
Q: Can I get more details on auto-healing / auto-resume of bots in case of failures due to application changes?
A: Generative Recorder, one of the new features that helps auto-resumption of the bots even in the case of application changes, is on the roadmap. We will be able to share more details in the coming months.
Q: Earlier in the slide deck there was a single line talking about Automation Co-Pilot for Automators generating documentation of a task bot. Could you preferably demo that, or at least speak to that output of that functionality and what that actually looks like?
A: We cannot demo that, as it’s in the works and on the roadmap. At a high level, it prdouces a high level summary of your automation, which you can then use as a reference for your change management steps.
Q: Is it your own LLM?
A: Yes, it's our own generalized LLM that we have built using our anonymized and aggregated metadata. This means that when you use Co-Pilot for Automators or Suggested Actions in your A360 control room, it's outputs are completely personalized and constrained within your own environment, and cannot be accessed outside of your environment. Again, this is explained in detail in our security/architecture documentation.
Q: For future session, but quickly becoming a cause for concern, third-party 2FA, MFA is starting to become a showstopper. Would very much like to know how the automation industry can work around this as it becomes more prevalent?
A: That's a great question because 2-factor authentication has become a norm. So definitely, the automation industry is catching up. And we may have already seen some improvements in terms of general security announcements. We've brought in 2-based security wherever possible, and that's going to be like prevalent in all the areas within automation. So we are working towards bring in all the security controls in place, all the types of authentication that can be added to a different sets of automation elements.
Q: Can we expect something in the future where we enter the requirement document and Co-Pilot already builds a skeleton of the automation where the developer can refine it in future?
A: Yes, this is currently under evaluation.