Accenture Melds Smarts And Wares With Nvidia For Agentic AI Push

Over the past two years, enterprises have tried to keep up with the staggering pace of the innovation with generative AI, mapping out ways to implement the emerging technology into their operations in hopes of saving time and money, increasing productivity, improving customer service and support, and driving efficiencies.

However, a new wave is on its way in the form of agentic AI, widely described by proponents as a “paradigm shift” in an AI industry that seemingly is in a constant state of change. That shift involves creating AI systems that can go beyond simply answering prompts or taking instructions to complete tasks to acting more autonomously – making decisions, developing plans, resolving problems, and taking action – with limited human supervision and to resolve problems that further the goals of the business.

Agentic AI systems comprise multiple agents that can have specific capabilities and can be orchestrated or chained to work together to complete complex tasks. They can plan, incorporate memory, use software tools, learn from their experiences, and adapt to changing circumstances.

“The opportunity for agents is gigantic,” Nvidia co-founder and chief executive officer Jensen Huang said during a conversation with Salesforce chief executive offer Marc Benioff at Salesforce’s Dreamforce show last month. “It sounds insane, but here’s the amazing thing: We’re going to have agents that obviously understand the subtleties of the things that we ask it to do, but it also can use tools and it can reason. It can reason with each other and collaborate with each other.”

Organizations can use agents to solve a complex problem and, if needed, those agents will find other agents to help, Huang said.

Developers for the past couple of years have been experimenting with ways to break down tasks into smaller pieces that AI systems could address autonomously and emerging frameworks like LangChain showed how agents could use code to interact with APIs. In recent months, IT vendors aggressively been bringing agents into their products and services. Most recently, that includes companies like Oracle for its Oracle Database 23ai, Salesforce with AgentForce, and ServiceNow in its Now Platform Xanadu release.

Cybersecurity firms also are introducing agents into their platforms and products to enable generative AI capabilities to not only detect threats and alert IT but to mitigate the threat rather than having to wait for a human to fix the issue.

This week, giant professional services company Accenture announced an expanded partnership with Nvidia that not only includes a new unit – the Accenture Nvidia Business Group – that has 30,000 people who will use AI agents to help enterprise scale agentic AI deployments, but also the Accenture AI Refinery platform that leverages Nvidia’s AI stack – Nvidia AI Foundry, AI Enterprise, and Omniverse – to make easier for them to adopt it.

Accenture also leverages Nvidia’s NIM and NeMo offerings for better token efficiency and for fine tuning and evaluation, said Justin Boitano, vice president of enterprise AI software products at Nvidia.

Accenture is making AI Refinery on both public and private cloud platforms and integrate it with its other business groups. There also will be network of engineering hubs to accelerate the use of agentic AI and Accenture is embracing agentic AI internally, initially with its Eclipse Automation business that it bought two years ago and within its marketing unit, which will lead to as much as 35 percent reduction in manual steps, 6 percent cost savings, and a 25 percent to 55 percent increase in speed to market.

Lan Guan, chief AI officer at Accenture, said during a briefing with journalists that generative AI demand drove $3 billion in bookings in its last fiscal year, and that agentic AI will help accelerate demand.

“The next wave of Agentic AI will be a huge step forward, will be a game changer in how our clients reinvent their business,” Guan said. “To me, it’s no longer just about prompting the prebuilt large language models and waiting for response. [With] what we call ‘zero-shot prompting’ right now with agentic AI … we can actually develop specialized cap that can independently and autonomously interact or act to make progress against goals or human’s intention. We are heading towards AI that does not just respond, but also actually learns, improves and work as a part of the team.”

Guan sees Accenture’s agentic AI comprising an “army of agents” consisting of three types. There are utility agents that specialize in a specific task – she used a research agent as an example – super agents that act as team leads activating utility agents, and orchestrator agents, project managers that orchestrate the various workflow functions across the enterprise.

“They seamlessly work together to make decisions and execute with precision across even the most complex workflows,” Guan said.

AI agents dovetail somewhat with the movement that’s been underway since the release of ChatGPT almost two years ago to make it easier for enterprises to customize AI system to their needs, as is done with retrieval-augmented generation (RAG), which organizations can use to bring their own corporate data into the training set for AI models.

Accenture’s AI Refinery includes not only prebuilt agents but also agent builders for creating custom agents, Guan said. Organizations also can build their own AI models using their own data and it has a “switchboard architecture to route between different models to pick the models that is aligned with your objectives without getting locked in,” she said.

“It can also integrate both internal and external data. This is this enterprise cognitive brain to integrate data to power enterprise-wide insights and provide this context engine knowledge engine to make a lot of responses and use of the agents out of the LLM [large language model] or multi-modal model very specific relevant to your company and to your industries.”

As with most everything in generative AI, there also are challenges when it comes to AI agents, including such standbys as data security and privacy, potential bias, ethical issues, and errors made by the system. As the AI systems become more autonomous, the need for human oversight and intervention will grow.

But that will have to be dealt with because the generative AI train is barreling forward and will only speed up as agentic AI comes on board.

Agentic AI is “a leap forward in the progression of the technology that is going to change the way we think about productivity and innovation in the context of scale AI,” Guan said, adding that “agentic AI will become a central piece of innovation within organizations and business.”