Meta, Outshift, Intuit and Asana delve into the future of agentic AI

While business and technology leaders are working today to realize the benefits of artificial intelligence, the future lies ahead and it is critical to build the foundations for what is soon to come. Technology leaders assume that the future of artificial intelligence will be agent-based. In other words, organizations will deploy intelligent systems that not only perform tasks on their own, but also make decisions with near-human precision, from writing code to handling e-commerce operations, acting as automated sales agents, and more.

Agentic AI was the focus of the VentureBeat AI Impact Tour: “Agentic AI – The Next Giant Step Forward in the AI ​​Revolution” presented by Cisco’s Outshift. Speakers from Outshift, Meta, Asana and Intuit joined VB CEO Matt Marshall to learn how organizations can plan for the future of agentic AI and other rapid advances in the field.

Building an agentic future

“If we think about a future where these agent systems work together to solve bigger problems, we need distributed agent systems and an open, interoperable Internet of Agents,” said Vijoy Pandey, general manager and vice president of Outshift by Cisco Marshall. “Innovation slows down when you’re in a walled garden. Whether you are an infrastructure provider, an operator, an application developer, and most importantly, a consumer or customer, an open system delivers value to every single link in the chain.

“Innovation slows down when you’re in a walled garden.”

Agent systems that learn how to talk and connect with each other have the power to transform the way people work, starting with software and IT and moving toward knowledge, service, and even manual labor as robotics evolves . They also need to be integrated with, and instantiated into, existing software systems and physical environments, whether cloud, on-premises, or embedded in a robotics solution.

Tying it all together requires layers of abstraction. This will look like open models, open tools, an orchestration and discovery layer, and then a communication layer that is secure, stable and open. Then we have probabilistic results, communication via NLP, exchange of state information and more.

“These could be some pretty huge problems to solve,” Pandey said. “We look for these problems, we look for what they look like and how to solve them. The future is there.”

It’s time to start using AI agents

The biggest question today is whether the technology is mature enough to realize its full potential – and it’s not over yet. But that shouldn’t be a barrier, Mano Paluri, vice president of AI generative engineering at Meta, said in a one-on-one conversation with Marshall.

“You can’t wait. Agents clearly feels like the next step in the evolution of these models.”

“You can’t wait,” he added. “In that sense, I would say he is ready. Agents clearly feels like the next step in the evolution of these models. The way we thought about it was that we were moving away from a model to a system consisting of many components that could be customized to suit our needs.”

In the pursuit of autonomous systems that can predict, learn, reason, act and iterate to solve a complex problem, we have already come far in perception – basic models can learn from text and images. We’re still in the early stages of solving complex problems, but today’s models can learn on a much larger scale than ever before in the last decade. These models begin to plan from both an inner and outer loop perspective. Today, the outer loop is the human training the model. Next will be the agent service parameters themselves.

Today’s Meta AI agent is the first step in the evolution of LLM as Meta moves away from a model towards a system composed of many configurable components. The goal is to tune the model for each use case, expand the context window, adapt to the new language, etc. for all four billion customers.

“We also believe in the agent family,” he said. “This incarnation of Meta AI is a user assistant, but we also believe that everyone should be able to customize the agent the way they want. This is an agent family where companies can create an acquirer. Creators can have their own agent to achieve greater scale. Advertisers may have unprecedented creative opportunities.”

Agentic AI use cases: Challenges and opportunities

To wrap up the evening, Paige Costello, Head of AI at Asana, Shubha Pant, Vice President of AI/ML at Outshift by Cisco, Kumar Sricharan, Vice President of Technology and Chief Artificial Intelligence Architect at Intuit, joined Marshall for a conversation Agentic artificial intelligence and the challenges and opportunities it will bring will emerge in application cases.

Real-world case studies

Handling requests through a workflow can take a lot of time, but that’s where agent-based AI comes into play. Asana has built-in agentic AI for both chat and workflow. For workflows, it can handle the request early on, determining where to prioritize it, whether there is enough information to start, and who should be included. This is a great place for a company to start adding agent-based workflows, Costello added.

“The question for the agent is how much decision-making or autonomy does he have to do these things in the context of these workflows?”

“There are many opportunities where artificial intelligence can be a partner in doing this work,” Costello said. “The question for the agent is how much decision-making and autonomy does he have to do these things in the context of these workflows? We have had great success in cooperation with security companies and marketing agencies. There are many other use cases where we’re starting to see things like creative requests, patch work, and feedback and commit loops.”

At Intuit, they automate the entire suite of financial products and offer financial insights and guidance. They are experimenting with artificial intelligence throughout this ecosystem of tasks, especially in areas where creating hand-crafted solutions would be time-consuming or even simply ineffective. For example, small businesses have a wide range of characteristics and needs. During onboarding, the customer is required to provide all of this information in detail so that Intuit can classify it.

“These agents essentially act autonomously based on this information and help the customer onboard with minimal effort on their part.”

“Now, with the development of agent-based AI, we are finding that we can use these systems, different agents that can work together to allow the customer to access different sources of their information,” Sricharan said. “Then these agents essentially act autonomously on that information and help the customer on-board with minimal effort on their part.”

Internally, agent-based AI helps the company navigate changes in tax laws that impact products. Instead of a team of developers researching and implementing new elements, they can use agent-based AI there to perform a variety of functions, from detecting where changes have taken place, to associating with our code and determining what changes to make next, acting as a co-pilot for developers.

“The goal is to predict IT problems before they occur, identify root causes when problems occur, and offer mitigation strategies until complete resolution.”

Outshift has a team of incubators dedicated to creating a multi-agent predictive diagnostics and repair tool for enterprises using various technologies. The goal is to predict IT problems before they occur, identify root causes when problems do occur, and offer mitigation strategies until complete resolution. Other agent-related AI projects are underway, including architecture development, open standards for agent orchestration, multi-agent systems composition, and an open agent protocol for inter-agent communication.

“The main challenges facing agent-based AI today are three-fold,” Pandey said. “First, how do AI agents discover each other and understand each other’s capabilities. Second, how do they collaborate to solve problems and deal with uncertain outcomes. And third, how they communicate using imprecise natural language instead of fixed structures like traditional APIs.

“We need to figure out how to create open source standards and guidelines for AI systems that take probability into account,” he added. “It’s time for the technology community to come together and build these solutions together.”

---

Sponsored articles are content created by a company that pays to publish or has a business relationship with VentureBeat and is always clearly marked. For more information, please contact us [email protected].