When Is It Going To Really Arrive?

Learn more about how Cognizant is using multi-agent systems via the Cognizant Neuro AI webpage

Every transformative technology rides a rollercoaster, first rising on a high of frothy expectations before sinking when inflated hopes are not met with sufficient speed. After a delay, there’s a dramatic bounce back up as the initial optimism is eventually earned with genuine success. Generative AI is no different.

Historically, it has taken at least a decade for any transformative innovation to mature. For the first 25 years of its existence, the Internet was limited to academia and military research. The browser arrived in the mid-1990s, but large companies did not take full advantage of it for another decade, when the Internet truly started to transform business.

The trajectory of the personal computer tells a similar story. In the early 1980s, PCs were used for word processing and spreadsheets, but they were not fully integrated into daily business activities. They acted as productivity assistants, but they did not transform enterprises. PCs were definitely not the indispensable business tools that they are today.

So, when will the transformative potential of Generative AI start to be realized at the enterprise level? The answer to that question can be found by examining the triggers that started the mass adoption of the Internet, PCs, and many other transformative new technologies. While the timelines may vary, the lessons remain the same.

Integrating a new technology into our lives is always hard, and it is even harder to do so at the large scale required for enterprise transformation. It requires an upfront investment of time and training to adopt new behaviors and construct new work routines. In order to make that investment, executives and users need to see a concrete payoff or existential risk that outweighs the powerful forces of inertia and fear of change.

Today, most people use Generative AI tools in a personal capacity more than they do for work within their enterprises. We believe the trigger to large-scale enterprise-wide adoption of Generative AI is the arrival of so-called multi-agent AI systems, which will unify the plethora of independent AI tools that exist today into powerful networks that can solve far more complex and valuable problems. Multi-agent systems provide the necessary payoff required to make the mass adoption of Generative AI worth it to users, executives, and companies.

Here’s why.

Today, many enterprises are incorporating the most widely available form of Generative AI, chatbots powered by large-language models (LLMs). This allows them to emulate human language quickly and seemingly naturally. For example, companies are integrating chatbots into their intranets for relatively simple employee use cases or using them as an entry point for customer service interactions. Deploying these chatbots to address specific applications across departments including HR, Finance, Legal, IT, and others has been significant for these businesses, but it has not been transformative.

This current approach is having limited impact because of one critical issue: users, including employees or customers, making inquiries through a top-level chatbot are often required to repeat their queries to department-specific tools.

But what if all those systems could interact with each other seamlessly? That is the promise of multi-agent AI systems. This approach would break down the silos between the various AI systems, giving the combined system game-changing potential.

Consider, for example, a multi-agent AI system at a manufacturing company. A sourcing agent would analyze existing processes and recommend more cost-effective alternative components based on seasonality and demand. This sourcing agent would then connect directly with a sustainability agent to determine how the change would impact the company’s environmental goals. Finally, a regulatory agent would oversee compliance activities, ensuring teams submit complete, up-to-date reports on time to reflect the updated supply chain. Instead of users having to take the information from one system and use it to feed another, the power of multi-agent artificial intelligence would allow all of these activities to take place with unprecedented speed, scale, and reliability.

The good news is that many companies have already begun to organically integrate LLM-powered chatbots into multi-agent systems. But it will be hard work, and there is no one-size-fits-all solution. To make optimal use of a multi-agent system, each enterprise will have to customize it for their unique data, processes, and culture. This will require ambition – using AI not only to improve existing processes but also to reimagine new ones. To effectively do this, organizations should start with their most important KPIs, the ones they really want to change, and work back from there. Starting with these crucial applications will ensure that the changes are embraced by employees rather than being forced on them. This comes with risk, however, and requires an okay-to-fail mindset that goes well beyond merely training people on the technology. It involves leading through changes and orchestrating the many parts of the organization into one powerful whole.

As ambitions rise to match the technology’s potential, businesses won’t merely be riding the AI rollercoaster, they’ll be enjoying the next upswing.