Cognizant: Is DeepSeek Really AI's Sputnik Moment?

When DeepSeek launched its R1 model at the end of January this year it caused many rival western companies’ stocks to plummet and led commentators to start asking what it means for competition, innovation and the future of AI.  

Since then it has made its older models open source and has propelled China back into the AI race, with other new AI agents like Manus being launched. 

However it’s worth taking a step back and looking at the bigger picture. 

DeepSeek’s model is remarkable. It can perform the same reasoning tasks as larger systems at a fraction of the computational power, but it is not the watershed moment it’s being hyped up to be. 

Instead of looking at comparisons to past breakthroughs like Sputnik, let’s look at what DeepSeek tells us about where AI is going. 

It is an important industry correction to accelerate AI innovation and adoption, and it’s one that’s been a long time coming. 

Democratisation is the future 

The first computers took up entire rooms and could handle relatively simple calculations. Today, our watches have more processing capacity by orders of magnitude. 

The driving force behind this evolution was not a fundamental change but rather the continuous optimisation of processing power, allowing it to be incorporated into smaller form-factors at a substantially reduced cost.

In a similar way, DeepSeek has used creative optimisation to make its model smaller and faster. This is a signpost of where the wider industry is going, and will be necessary for enterprise-grade AI to gain mainstream adoption. 

Traditional, larger AI models have been slow and expensive to run, which limits their usage potential. 

By comparison, smaller DeepSeek models tested on a MacBook were much more powerful than similarly-sized models of just a few months ago. No massive processing power or storage was needed. Additionally, these models can be trained faster while requiring less horsepower.

The takeaway here is to look at where it leads us: democratisation. The exclusive ownership of powerful models by a few commercial entities is fading. This is good for us all. 

DeepSeek creates space for a multi-agent world 

Until now, only a few AI models have met the threshold to support advanced AI reasoning capabilities and can work as acceptable large language mode (LLM)-based agents. 

Now, DeepSeek has cracked the door open and can inspire more models to achieve the necessary results without the need for LLM models. This could have a fundamental effect on agents and multi-agent systems. 

We’ve all seen how companies are turning towards agentic AI as a powerful way to leverage AI, and we are starting to see how individual AI models can be connected into multi-agent systems that deliver enormous gains in productivity and autonomy. 

The optimisation improvements DeepSeek has introduced allow enterprises to consider a wider scope of use cases for multi-agent systems because we can now begin working with cheaper, smaller and faster AI models.

As these new models begin to take hold, businesses will gain far more flexibility in where and how they build and deploy them. 

In turn, this can significantly accelerate the AI adoption timeline many enterprises are on.  

Expanding the potential use cases 

As LLMs become faster and more energy-efficient, the feasibility of multi-agent solutions increases. Consequently, a broader range of use-cases will benefit from multi-agent applications. 

So far, use cases have focused on where a decision maker within a certain domain uses multi-agent systems to augment process flows among a few users. 

The challenge comes when enterprises try to scale these solutions to thousands of people using LLMs. 

Under load, these models can be slow and inefficient in multi-agent settings. But now organisations can use cost-effective, smaller and faster models to run systems at scale and increase the number of people who use them. As the types of process flows and throughput capacity increase, use cases proliferate.

The most obvious area to immediately benefit would be call centre and support line augmentation, with multiple AI agents handling a customer inquiry from start to finish.

Another example is the insurance industry. A property underwriter typically assesses large amounts of third-party data to help them understand past instances of underwriting similar property, and conclude if and how to ensure. 

This is a complex process, but as multi-agent systems become more accessible, an insurance company could augment this process through interconnected agents. These agents can analyse third party and internal information, distill and consolidate perspectives, and provide various options along with predicted outcomes for the property that account for risk, win-loss, and cost. 

What will the future bring?

AI development has focused on building LLM-based agents which can act as one-stop-shops for everyone’s needs. 

However, this model is hard to scale, and ‘bigger is better’ is not always the best way forward. Instead, harnessing the collective intelligence of a multitude of smaller, more cost-effective AI agents that are smart enough to represent various process nodes in a business organisation flow could provide a better solution.

DeepSeek has rightfully inspired the industry to embrace models that are faster, more efficient and more cost-effective. This brings significant promise to what can be achieved through cross-enterprise use case deployments. 

As the dust settles around DeepSeek, I’m hopeful we’ll start to feel the early rumblings of a gold rush in terms of what these models can be used for, and how enterprises can harness this power to deliver breakthrough transformation. 

Explore the latest edition of AI Magazine and be part of the conversation at our global conference series, Tech & AI LIVE. 

Discover all our upcoming events and secure your tickets today.

AI Magazine is a BizClik brand