1. Inquire about the future roadmap for Automation Anywhere. Request an overview of upcoming features, new AI Agent functionality announced for v38, MCP server, A2A and document extraction by Agent on the fly, etc.
2. Please provide a detailed technical breakdown of how New AI agent performs tool selection.
3. AI Agent testing and Error handling : How can we effectively test and manage errors for an AI agent in the UAT environment, to build confidence that the entire process will operate smoothly and reliably in production?
4. How to calculate ROI of APA?
5. Are you going to replace document automation with Agentic AI in future as LLMs can also extract data from file?

Hi ​@Vineet Pujari 

I was an UiPath dev for about 3.5 years and switched to AA in my new job. One of the things I noticed here is, community itself is very small and usually unresponsive. In UiPath forums, we get incentives when we answer the questions every month (For about top 20 - 30 people), why not have a similar initiative here in AA, which encourages people to participate more in the forums and grow the community?

Hello ​@Vineet Pujari,

LLMs are overkill for 80% of business tasks. Enter Small Language Model 

Unlike large language models that need tons of computing power, SLMs run on fewer resources. This means smaller companies, individual developers, and even startups can use them without needing massive servers or huge budgets. what is your thought on it?

Thanks ​@Pravin 3017 for a very relevant question in today’s time when the investment in AI models must justify the business outcomes they produce. I think I would always prefer to start with the smallest model that meets your quality bar and your ROI bar. You can scale up only when the data says. In a production scenario, I would recommend a hybrid approach. Let an SLM do 70-80% of the requests and escalate only the harder ones to an LLM.
We demoed one of the SLMs in a product club sessions in the past by creating an SLM connector using the Automation Anywhere Connector Builder. Do give it a try and let me know your experience.

Thanks ​@Pravin 3017 for a very relevant question in today’s time when the investment in AI models must justify the business outcomes they produce. I think I would always prefer to start with the smallest model that meets your quality bar and your ROI bar. You can scale up only when the data says. In a production scenario, I would recommend a hybrid approach. Let an SLM do 70-80% of the requests and escalate only the harder ones to an LLM.
We demoed one of the SLMs in a product club sessions in the past by creating an SLM connector using the Automation Anywhere Connector Builder. Do give it a try and let me know your experience.

​@Vineet Pujari 

I have created aari process. In that aari process I am just calling a simple task bot (Update Task). Now when I am running this AARI process directly by opening the AARI Process then this Update task is getting called.The same is displayed in AARI request as whenever we ran this aari process request gets created.

Now issue is when I am running the same AARI process using Created request command in one task bot (Create Task) then request is getting created but this Update Task bot is showing unsuccessfull. In audit logs and in activity this unsuccessfull is not

     OR 

can we call the process via API

As Automation Anywhere evolves with Agentic Process Automation (APA) and integrates more AI capabilities, I'm thinking about the fundamental components that will drive this new era. My question is about the role of API Tasks in the APA journey. I believe that API Tasks are more critical than ever for a successful APA implementation. Without them, the pace of automation would be significantly slower. API Tasks offer a more direct, resilient, and faster way to connect systems. This seems essential for the dynamic, real-time decision-making that AI agents in an APA system require.

Why do you believe API Tasks are more critical than ever for a successful APA implementation, and how do they accelerate the transition from RPA?

I’ll try to answer these succinctly as provide links and resources for more details:

We have quite a few game-changing features lined up in A.38 and the following releases. You correctly stated some of them. Our own MCP server that lets you consume automations built on Automation Anywhere, from your models. You should definitely attend our upcoming sessions where we cover the A.38 features going live very very soon.

Doc Auto is integral to our Agentic Process Automation platform and already leverages document specific models and LLMs to power extractions. Its input and output seamlessly integrate with any agent which is part of our customer's "agentic process"

​@nirmal.mathimaran Great question especially because API Task is so close to my heart our relationship has zero downtime :)
Here’s a short answer: Because AI Agents need “tools” and API Tasks are the only tools with guarantees. They give you the reliability, speed and a great user experience where as UI bots give you pixels and hope.
And a longer one:
(a) When an AI Agent needs to present information/result to a business user, the expectation is to present it in a few (milli)seconds, while the user is looking at a form. This is possible only via API Task.
(b) When an AI Agent determines that it needs to deploy multiple “tools” in parallel so as to determine the next course of action, API Task can spin up additional cloud runners instantly. Without the auto-scaling API Tasks, you are left at the mercy of limited VMs on-premise that are going to take longer to process the workload and will not be able to maintain SLAs.
(c) API Tasks are totally safe w.r.t handling your data. No data gets transferred or stored on the AA cloud. Only the package metadata (you know, the commands and the logic) gets into the cloud in an encrypted manner. The cloud runner is automatically scaled-down ie destroyed when there are no upcoming executions. Plus compliance with frameworks like GDPR, CCPA and SOC2 should give you the required assurance to run your workloads in it.

Bottom line: move everything you can to API-first tasks, keep UI bots only for the true long tail. In this way, your agents will be faster, cheaper and safer with a clean transition from RPA to APA.

Thanks for the fantastic answer, ​@Vineet Pujari ! The points you made about AI agents needing "guaranteed tools" and the distinction between "pixels and hope" and "reliability and speed" really resonate.

I especially appreciate the breakdown on:

• Speed: The ability to present results in milliseconds, which is crucial for a positive user experience.
• Scalability: The power of instant, parallel processing and auto-scaling cloud runners, which directly addresses my next point.
• Security & Compliance: The emphasis on data safety and compliance, which is a critical consideration for any enterprise.

This brings me to my next question, which directly follows from your point about scalability and safety.

How do you see these advantages in scalability and resilience playing out in real-world scenarios? What are some concrete examples of where API Tasks have been a game-changer for large-scale, enterprise-wide automations, particularly in the context of an APA framework?

This also makes me wonder about the flip side of the coin.

Are there any specific scenarios or edge cases where the API-first approach might be a limitation, and traditional RPA bots are still the best, or even only, option?

I'm interested to hear about both the successes and the challenges in moving away from RPA automation.

Looking forward to your thoughts!

Hi ​@hardik.choudhary,

Yes, a processes can be submitted in a number of ways:
(a) By a user
(b) By a bot via the Create Request command in the Process Composer package
(c) Via the /requests/create public API 

I think the problem you are running into can be fixed by following these steps:
(1) Default bot runner device - Make sure you have a device available and it is either the default device or is part of a device pool
(2) Enable the process requests to be created by a bot - This setting can be found in Manage > Processes > Your process
I’ve attached a zip file here with the video that shows what I explained here.
Let me know if this helps fix the issue.

​@Vineet Pujari  Thanks a lot.Resolved