Agentic AI: a problem of scale

Markus Muller at Boomi explains how to prevent agentic AI sprawl from turning into chaos

 

For years, organisations have struggled with application, data, and API sprawl. Every new connection to their IT systems has exposed enterprises to potential cyber-security gaps and increased non-compliance, reputational, and financial risks. With the advent of agentic AI, these issues will be compounded even more unless organisations have the right controls in place. 

 

 

AI and data everywhere

Every AI agent created is another connection to enterprise data that needs to be managed. If organisations start using multiple AI agents to help employees make decisions and assist with their daily tasks, we will soon live in a world where a single enterprise is managing millions of agents. As this trend takes off, Nvidia CEO Jensen Huang claims that it will one day be “a 50,000-employee company with 100 million AI assistants”. 

 

Without automated processes to optimise and secure all these agents, organisations risk losing control of their data. To prevent this, IT teams must ensure they future-proof their AI management processes. This means gaining oversight of the development, deployment, and governance of AI agents, ensuring they are integrated into business and technical workflows safely and responsibly. 

 

But how do enterprises build a strong AI management strategy that can scale as rapidly as their agents?

 

 

Easy access saves enterprises pain

Enterprises should first focus on making it easy for AI agents to draw frequently used internal data from a trusted source. By ensuring most of the data that users will need is at their fingertips from the outset, there will be less need for users to toil away trying to connect agents to the source they need, saving valuable time and minimising potential compliance risk. 

 

Ensuring easy access to common data sources that are approved for use will also help to enforce safe and ethical behaviours when using AI agents. This is because the enterprise can set guardrails that promote privacy and security best practices, instead of relying on users to connect AI agents to data sources that might include sensitive personal or company data that shouldn’t be shared. With an easy route to start connecting their AI agents to the data that powers them, users will be able to generate more reliable results, all in a secure manner. 

 

 

Bringing AI agents into focus

Even with the right guardrails in place, enterprises must achieve “always-on” visibility into the AI agents in use across the organisation to maintain security and ensure the quality of outputs. To enable this, IT teams need the ability to automatically discover AI agents when they are introduced to their environment, and see exactly when and how they use APIs to interact with systems and access enterprise data. 

 

This will help enterprises to understand AI’s impact across the IT ecosystem, identify exactly where it’s been used, analyse the quality of the output, and detect any potential security or privacy threats. That won’t just reduce risk, it will also help organisations to identify new ways to optimise their use of AI tools. 

 

With the ability to automatically pinpoint failures and identify opportunities to optimise their agents, organisations can also continuously improve the accuracy of their AI and enhance the experience for the employees who use it. This capability will also reduce compliance risk when greater calls for AI transparency and explainability come into force, either through the EU AI Act for high-risk use cases or later regulations.

 

 

Taking a holistic approach

To effectively manage AI agents at scale, enterprises must also be able to create them without constantly re-inventing the wheel. The most effective way to support this is by creating a user-friendly environment where employees can design and deploy their AI agents using repeatable low-code templates and a natural language interface that doesn’t require advanced knowledge of AI. 

 

By creating a centralised hub for AI management, organisations can create AI agents straight out of the box, or they can repurpose and customise existing agents for more specialised use cases. But critically, each agent created will utilise what’s already been built, ensuring best practice and helping to maximise the time to value for the user whenever they deploy an agent.

 

What’s more, with this unified approach, organisations can more easily implement agent lifecycle management controls to conduct continuous checks that make sure systems are integrated effectively and remain compliant with the latest regulations.

 

 

Sending agents into the field

As AI agents proliferate across every department – from IT to human resources and marketing functions – organisations must deploy a full AI lifecycle management strategy where their applications, data, and API integrations can be managed holistically. This is the only way to ensure employees can unlock the power of AI quickly and effectively, without sacrificing security, privacy, or efficiency.

 

Once enterprises can create agents capable of removing the strain of repetitive workloads in an instant, the sky is the limit. Well-managed AI agents will help employees supercharge productivity, drive innovation, and pull ahead of the competition in an increasingly automated future.  

 

---

 

Markus Muller is Global Field CTO at Boomi

 

Main image courtesy of iStockPhoto.com and Andrey Suslov