DeepSeek and the future of enterprise AI agents

The enterprise AI lag

With the global AI market predicted to reach $826bn by 2030, it’s understandable that organizations are looking to leverage the opportunities it promises. However, we’re seeing large enterprises struggle to adopt AI, largely thanks to their inability to pivot and execute with the speed and adaptability of more agile scale-ups.

Unily's recent study, the AI reality check, uncovered how well large enterprises were faring at implementing AI. We gathered the views of 800 enterprise knowledge workers, to gain a clearer picture of what their employers were and weren’t doing. As employees are at the heart of the AI conversation – and the ones who will ultimately deliver the pace of change enterprises require – getting their insight is essential.

One of the most eye-opening revelations was that almost a third (32%) had never used an AI tool at work, while 40% think their enterprise is falling behind competitors because of slow AI adoption.

This highlights that simply acknowledging the impact of AI isn’t enough to drive adoption. The key lies in providing employees with the tools to adopt and implement AI into their daily work lives.

AI agents, the enterprise game-changer? 

Transformation in AI for large enterprises can be found in AI Agents. AI Agents are more action-orientated, they’re autonomous systems which can plan, optimize and execute complex tasks – an obvious choice for unlocking productivity gains.  

Despite the vast amount of AI Agents available in the market, many enterprises are still grappling with which agent to partner with.  

We are also seeing a trend with our more AI mature customers who are investing in training Large Language Models (LLMs) to give them an edge on their competition, often ingesting proprietary data that can then power AI agents. While traditional AI agents perform tasks based on predefined rules and inputs, Agentic AI takes this a step further by working towards specific goals, without needing human intervention. Agentic AI acts within the context of its environment – able to make decisions to achieve specific objectives. This makes the agents digital co-workers, rather than just automations. The benefit for enterprises is being able to free up their human talent for higher-value work. 

Moving forward with multi-agents 

Part of the reason DeepSeek has been making waves recently is because it claims to provide a faster and more cost-efficient way to build AI agents. This could be particularly appealing to enterprises who want to build their own models and agents, rather than using 3rd party tools. 

For example, Finance departments need a particular AI agent for automating invoice and expense payments, while HR departments use one for things like scheduling interviews. 

However, even within these individual departments, there are many different use cases. If we use the example of HR again – a multi-agent system would see one handle job application screening, another manage interviews, and another assist with onboarding. Rather than operating independently, these agents collaborate with each other to achieve tasks. 

A challenge this creates is finding a way to manage and keep track of each agent. Recently, Workday launched a platform to help enterprises manage a fleet of AI agents in a central place. Solutions like this will be the way to go for enterprises – not only giving them greater control, but increasing adoption by reducing complexity. This approach to multi-agents is one that Unily announced at their flagship event, Unite 2024 in October.  

Ungoverned AI is unadopted AI 

Our study pinpointed some specific areas where enterprises need to do more to drive adoption. One of the main reasons is a lack of governance. 52% of employees say their organization has no AI policies, and 40% say they’ve never received any AI training.  

The lack of clear policies is leading to a major disconnect. On one hand, business leaders are convinced about the transformative potential of AI – with a Gartner study revealing 74% of CEOs think it will significantly impact their industry. However, the wider organization feels ambivalent – with a sense that integrating AI is not actually a key priority. In fact, 20% of employees currently fail to see the business application for AI, showing that more work needs to be done on aligning them with leadership. 

It’s also important to note that there’s still an element of the unknown about AI, which naturally leads to uncertainty for some employees. By not being given clear direction from the top, this cohort will continue to hesitate about adopting AI – meaning the business will continue to lose out on its benefits.   

The risks of employee curiosity  

Ungoverned AI isn’t just about poor take-up rates. In fact, a significant proportion of employees – 27% – say they’ve taken it upon themselves to adopt AI technologies, with productivity consequently jumping by 25-90%. This sense of curiosity and ambition might seem like something to applaud, but without guidance from leaders, there is a huge risk of AI becoming shadow IT. 

This refers to employees using unauthorized software, apps or tools without the knowledge or approval of the IT department. It’s something which has been a creeping issue for a number of years – a 2023 study by Devo saw 96% of IT security professionals admit to someone at their organization using AI tools not provided by their company. 

With the vast range of consumer-grade AI tools available – from productivity apps to generative writing assistants – the appeal for employees is obvious. However It comes with clear dangers for enterprises, such as: 

• Data leaks 
• Cybersecurity threats 
• Compliance risks in regulated industries 
• Duplicate costs 
• Excessive Digital Noise 

A study by data breach firm Hayes Connor showed that 11% of information pasted into ChatGPT was confidential. Our own research went along similar lines – with 9% of enterprise employees admitting to inputting sensitive information into non-sanctioned AI tools, and a further 11% unsure if they have. 

This makes it essential for IT leaders to implement Governed AI – not just to drive adoption, but to decrease risk. 

The importance of integration 

Governance isn’t the only issue enterprises are facing with AI agents. Just as challenging is being able to integrate them into existing workflows. Until this is achieved, AI agents will simply become a series of siloed tools – adding to the very same inefficiencies they are supposed to solve. 

Learning how to work with yet another tool is often regarded as a headache – and when organizations have multiple agents, they multiply this headache. The only way around this is to embed it into employee workflows and provide easy access through a centralized experience layer - an employee experience platform (EXP). 

In fact, our study saw 56% say they’d use AI more if this was the case. They don’t want more tools – they simply want the ones they currently use to be modernized.  

The benefits of true integration are clear – it streamlines processes, delivers faster knowledge and boosts productivity. 

How IT leaders are choosing AI agents 

Choosing an AI agent is very much dependant on organizational needs and ecosystems. There’s plenty of options out there, but here are some that IT leaders are commonly selecting. 

Microsoft Copilot: For enterprises that are already deeply integrated into the Microsoft 365 ecosystem, Microsoft Copilot is an obvious choice. Designed to seamlessly work with applications like Microsoft Teams, Word, Excel, Outlook and others, Copilot leverages Azure AI and OpenAI models to enhance collaboration and productivity. Some of the main productivity enhancements include the ability to summarize emails, draft reports and automate workflows. 

Gemini provides AI-powered enhancements that help teams collaborate more efficiently. Similar to how Copilot works well for companies already integrated in the Microsoft workflow, Gemini does the same for those in the Google ecosystem. It provides native integration across the full suite of Google tools (Docs, Sheets, Gmail and Meet), while using its cloud-powered AI for enhanced security. 

Moveworks is a powerful AI-driven assistant designed to streamline IT and HR support. Instead of relying on traditional helpdesk tickets, it automates responses to common employee queries – for example password resets and leave requests. For IT leaders looking to reduce the strain on their helpdesk and shift to a self-service approach, Moveworks is a good option. 

Workgrid: For organizations that need more flexibility, Workgrid allows them to create their own AI-powered assistants. These can be tailored to specific business needs – and crucially, there’s very little coding involved. This is ideal for enterprises that require custom AI workflows but don’t want to invest in full-scale AI development. 

Custom ChatGPT: Staying with the theme of flexibility, a Custom ChatGPT is great for businesses wanting a bespoke AI model. It can be created to fit your exact business and industry needs – for example a chatbot that responds to customer queries. While Custom GPTs do require more development time and costs, they offer maximum control and flexibility. 

The rise of Deepseek 

We’ve mentioned the buzz around DeepSeek and there are two main attractions for enterprises.  

While the exact numbers are not yet confirmed, it’s believed that the DeepSeek hardware is 20-50 times cheaper than OpenAI. Just as importantly, as an open-source model, it’s available for anyone to copy, download and build on (something that key players like ChatGPT have moved away from).  

This makes AI more cost effective and accessible for businesses. Crucially, the open-source element provides flexibility for them to develop and iterate their applications.  

The disruption DeepSeek has brought to the sector also brings another advantage for enterprises. There’s every possibility that this will now spur other companies on to compete – ultimately bringing more options to the table and driving costs down further. 

It’s worth noting that it’s still early days for DeepSeek as an enterprise option, so organizations should be wary about committing to what seems like the next big thing. In particular, there are still safety and governance concerns to be ironed out.  

Over the last few weeks, DeepSeek was found to have left more than a million lines of data unsecured, while a security investigation by Cisco found that it had a 100% attack success rate, failing to block a single harmful prompt. 

However, as enterprises demand more customized and intelligent AI solutions, there’s no question that DeepSeek could become an essential part of their toolkit.  

A New Way Of Choosing AI Agents 

When selecting an AI agent, or multiple agents, IT leaders must consider a number of factors, including: 

• Governance: Does it meet your organization's regulatory requirements? 
• Ecosystem fit: Does it integrate with your existing tools? How do your teams access it?  
• Customization needs: Do you need a plug-and-play AI or a bespoke model? 
• Flexibility: Can you avoid being locked into one default system? 

Going forward, the best approach for enterprises will be a mix-and-match one. After all, there are always multiple use cases for an AI agent(s), so it makes sense to use different ones for different functions.  

However, the best way to ensure these are adopted – therefore allowing enterprises to truly reap the benefits – is by linking them all via a single platform. Otherwise, they run the risk of creating digital friction in the form of siloes, multiple logins and constant app-switching.  

This is the exact opposite of what employees want as well – over half (51%) of our survey respondents wanted to see AI used to simplify the digital workplace, cutting through the complexity of fragmented tools and systems. Meanwhile, 34% feel overwhelmed by too many apps and tools – underlining the importance of ensuring your agents work with what your people already use. 

For example, many of Unily’s enterprise customers benefit from our world-first “bring-your-own-AI-Assistant” capability. Unlike our competitors who lock businesses into a default assistant, Unily has deliberately chosen a flexible and AI agent agnostic approach.  

This enables businesses to integrate their own Large Language Models and Digital Agents directly into the Unily platform. As a result, they can meet their employees where they already are and deliver a seamless experience. 

The key takeaway

AI agents are an area of rapid growth, and the opportunities they provide for enterprise efficiency are huge. However, they will only happen if the agents are adopted in a carefully governed manner. As our research pinpoints, adoption and governance are the two primary reasons enterprises are lagging when it comes to AI. 

The key is to recognize that it’s both tech and talent which come together to power organizational velocity – so it makes sense for an EXP to be seen as the enabler of your AI agents.  

A high-quality EXP will already have the infrastructure and ability to streamline workflows, engage employees and create a seamless end-to-end experience. Aligning this with your AI agents will ensure they get embedded into your employees’ day-to-day. This ultimately drives fast and sustainable adoption – helping your enterprise reap the true benefits. 

Start our Exclusive AI Product Tour