How Do You Manage AI Agents Before Your Tech Stack Spins Out of Control

AI agents are taking the world of business technology by storm. There’s a huge amount of hype and excitement surrounding these intelligent, always-on virtual assistants that promise to free humans from routine and repetitive workflows.

But as with anything new and exciting, there’s also plenty of confusion. In the past year, literally hundreds of platforms and providers have emerged.

Unfortunately, most businesses probably won’t find a “one-size-fits-all” solution to all the opportunities open to them.

This means they can end up with a stack of tools and applications from different vendors, all competing to sell tools that provide different methods of achieving similar goals.

So, businesses need to take practical steps to ensure their agentic ecosystems remain cohesive, secure and stable, as they expand across a growing number of use cases, business processes and teams.

Let’s take a look at how they can start approaching this.

Why Agentic Ecosystems Quickly Become Chaotic

The purpose of agents is to automate the routine and repetitive work that’s a part of just about every business function, from marketing and finance to HR, logistics and operations.

But when every team is building out its own agents, the agentic ecosystem can quickly become a confusing mess of duplicated work, incompatible standards and inconsistent workflows.

Different platform vendors like Google, OpenAI, Salesforce, Microsoft and Hubspot all work to their own conventions and standards. And that’s before we even get into coding custom or bespoke agents for niche use cases.

There are a number of dangers here: no centralization of permissioning, authentication and security. A lack of traceability and ownership. Unclear audit trails. No single source of truth for policy and governance. And potentially large amounts of time and effort are wasted on duplicated workflows.

All this can quickly become a recipe for confusion, security risks and plenty of administrative headaches. So, what’s the answer?

Open Standards

One solution to this fragmented approach lies in the adoption of open standards. Organizations like the Agentic AI Foundation (part of the Linux Foundation) are already working on this. Backed by OpenAI and Anthropic, along with numerous cloud providers, the AAIF hopes to do for agentic tools what the Linux Foundation did for open-source operating systems: develop universal, shared standards.

This means that agents will have standardized ways to connect and communicate with tools and services, as well as native layers for processes like security, authentication and ownership.

It’s designed to be compatible with agents.md, the open source markdown framework developed by OpenAI for specifying agent capabilities, behaviors and guardrails.

As agentic infrastructure matures, we’re likely to see further development and adoption of common, open standards designed to help organizations administer complex, growing ecosystems of agents.

Understanding and keeping track of emerging standards like this will be important for anyone responsible for implementing and maintaining agentic systems in the coming years.

Managing Your Agentic Ecosystem

In the meantime, as we wait for standards to settle, we need to start by ensuring there is clear ownership and governance. Here are some steps every business should take:

Establish a multi-departmental group with responsibility for approving, overseeing and reviewing use cases as they are identified by individual departments and specialists.

Create and adopt an acceptable agentic AI use policy. This should clearly define the ground rules around building, deploying and maintaining agents within the organization.

Maintain a centralized record of how and where agents are deployed, clearly labelled with the purpose, permissions and ownership of every agent. As well as assisting with administration, this serves the strategic purpose of ensuring agents are always aligned with business goals and priorities.

Develop, adopt and document standard operating procedures and best practices around agent design and workflow. This ensures the business grows more mature in its agentic capabilities as it learns in a structured way from each initiative. It also helps avoid wasted effort due to redundancies and duplicated workloads.

Implement auditability, transparency and traceability into your agentic ecosystem from the ground up, ensuring every action taken by agents is logged and can be traced back to the human or business function with responsibility for it.

Finally, review the performance, safety and relevance of every agent regularly. Any that have become redundant or simply aren’t being used should be retired promptly to maintain tidiness and avoid unintended security implications.

The Future Of Manageable Agentic Infrastructure

As you begin building your agentic ecosystem, what starts out as a handful of proofs-of-concept and quick wins could quickly evolve into dozens, or even hundreds, of autonomous systems embedded into core business processes.

Adopting a disciplined approach and following basic rules like those covered here is essential if you want to avoid creating confusion, risk and wasted time.

Of course, it’s early days, and other considerations and priorities are likely to become apparent as agents take on more responsibility across businesses.

One thing that will remain constant, though, is that organizations that successfully transition to the agentic era will be those that implement a layer of coordination and oversight rather than building in an ad-hoc and disconnected way.

If managed well, agents have the potential to unlock huge savings of time and resources. If not, those gains could quickly become outweighed by complexity, wasted effort and risk.