Many enterprises have already begun to explore AI agents and determine if deploying them would be a viable option for the business. Many still equate agents to something that most companies have been using for years: automation.

Automation pioneer

UiPath’s view of agents and orchestrating an entire ecosystem is a little different.

UiPath announced its new UiPath Platform For Agentic Automation. It was made clear, however, that agents are not an entirely new version of robotics process automation (RPA), but rather another tool that enterprises can use to complete workflows.

Daniel Dines told VentureBeat, UiPath’s founder and CEO, in an interview, that agents as they are currently built cannot be fully automatized. Dines stated that the current LLMs are nondeterministic and therefore cannot be run in an autonomous manner. If you look at the majority of implementations of agents, they are actually chatbots. We’re moving away from chat in and chat out to an agent which is data in and action out. This is where we orchestrate agents, humans, and robots. It manages the flow from the agents to the human employee and then to the automation layer. UiPath describes Maestro as an automated supervisor who “automates and optimizes complex business process” and monitors the performance.

Breaking down automation and agents

Maestro breaks down user prompts into manageable steps. Dines explained that Maestro does not allow agents to access information randomly. Instead, it has three steps.

1. The agent first analyzes the prompt and makes a recommendation on how to complete the query. The recommendation is then approved by a human user.
2. An RPA tool will then execute the recommendation, completing your request.

Dines stated that Maestro makes workflows more transparent and accountable, because a human is still in the loop while a rules-based RPA completes the task. UiPath separates agents that collect data and make recommendations from the automation that follows up on that recommendation. This ensures that enterprises do not give agents unfettered access their entire system.

I think it’s an effective way for enterprises adopt agents. Dines stated that in many conversations with clients they have expressed concern about the unlimited agency of agent. UiPath integrates with LangChain, a provider of orchestration frameworks to offer multi-agent frameworks. The Platform for Agentic Automation works with Microsoft and Anthropic frameworks. UiPath is part of Google’s Agent to Agent protocol.

Not all agents are automation

Dines insists on thinking about agents as a full stack solution, where agents read the data, and then take actions,

“Agents, being nondeterministic by nature, are transactional; They create effects on the systems underlying them. Dines said that no client he knows would take a risk on this. Automations are the only way to guarantee 100% reliability in transactions. Our solution is the best of both worlds.

Then he added that “maybe someday” agentic AI will “become more reliable and you can delegate some actions to agents, but there should be a progression.” Other industry professionals believe that agents are a next evolution in automation. Agentic AI is based on the idea that a system can act on behalf of the user. Many people want “ambient” AI agents that run in the background and act proactively for the user, alerting them to any changes they need to be aware of. UiPath must still prove that its approach is more effective than the all-in-one agents and that it cuts through the hype around agents that do everything for their users.

Companies such as ServiceNow, Salesforce and Writer have all released agentic platform aimed at enterprise users. Writer’s platform is based on self-evolving agents.

Enterprises were also excited about the idea that AI agents would streamline their work and automate manual tasks within companies.