How to prepare your business for agentic AI adoption

1. Identify key agentic AI use cases for your business

Part of the value of AI agents lies in their ability to accelerate virtually any type of business process, with customer service and software development currently among the more popular use cases for agentic AI.

But it's unrealistic to deploy agents across all business domains at once. Instead, agentic AI adoption should start with deciding which use cases or business functions to prioritize. Business leaders must determine where AI agents will add the most value based on which processes are currently inefficient or where the business lacks staff.

For example, if the finance department is understaffed, deploying AI agents capable of automating tasks like invoice management might be an optimal use case to prioritize. If the business is experiencing poor cybersecurity outcomes, focusing on deploying agents to automate processes such as threat detection and assessment might make sense.

2. Assess data readiness

AI agents are only as effective as the data sources that power them. Without relevant, high-quality data, AI agents won't be able to take effective action. An agent designed to automate customer service interactions, for example, won't work well if the business deploying it can't connect the agent to databases containing customer information and purchasing history.

Therefore, leaders must assess their organization's readiness for agentic AI adoption from the perspective of data readiness. Specifically, businesses should consider the following:

• Which types of structured and unstructured data AI agents can access within specific business domains.
• Whether data contains missing information, redundancies, inaccuracies or other problems that contribute to low quality.
• How to ensure that the most recent data remains accessible to agents as data sources evolve.

If your company's data is not ready for agentic AI, invest in improving the data before implementing AI agents.

3. Define limitations for agentic capabilities

Only a small fraction of businesses -- 6%, according to analysis from a recent Harvard Business Review study -- say they trust AI agents to carry out tasks completely on their own. Understandable because AI agents can make mistakes, such as accidentally deleting critical data.

For this reason, businesses should establish clear criteria for what they will and won't allow agents to perform. These rules might vary across business functions or use cases. For example, agents might be allowed greater autonomy when performing lower-stakes tasks like documentation generation, whereas a business might decide not to entrust AI agents with higher-risk processes in a domain like finance.

4. Decide when to require human review of agentic actions

Determine when to implement human-in-the-loop controls for agentic processes. These controls require humans to review and approve proposed AI agent actions before the actions take place. Not all agentic workflows should require manual approval, but it's appropriate for processes that present a certain level of risk to the organization.

A business, for instance, might not let AI agents complete financial transactions exceeding a certain dollar threshold until a human signs off but allow agents to carry out smaller-value transactions without manual approval.

5. Identify and monitor AI agents

When dozens or hundreds of AI agents are operating within IT systems, keeping track of which agents exist and what they can (and can't) do can become quite challenging. To address this issue, businesses must invest in tools that automate the process of identifying AI agents and tracking which actions they should carry out.

A key tool for this function is an agent mesh -- a platform that operates as an intermediary among agents, data sources and AI models. By detecting AI agents and monitoring the data sources they access and the actions they perform, agent meshes play a crucial role in scaling and automating agent identification and monitoring.

As McKinsey & Company defines it, agent meshes function as a "composable, distributed and vendor-agnostic architectural paradigm that enables multiple agents to reason, collaborate and act autonomously across a wide array of systems, tools and language models -- securely, at scale, and built to evolve with the technology."

6. Mitigate agentic cost-management risks

AI agents can be expensive to develop. And they can cost even more when factoring in the price of the prompts they send to AI models. To keep AI agents from eating up enterprise budgets, it's critical to establish cost-control strategies early in the planning stage.

One way to reduce development costs is to use open source AI agents or frameworks, which are generally free. In addition, agent meshes can help manage agentic AI operating costs. They can intercept communications between agents and AI models and strip out unnecessary data to reduce prompting costs.

Some meshes can also redirect prompts to lower-cost models -- a smart way to save money, given that the cost per prompt (or more specifically, per token, the units of data that make up prompts) can vary significantly among different models. For simpler queries, a lower-cost model might perform just as well as the more expensive one that an agent initially attempts to use.

7. Assess agentic AI vendors

For most businesses, it's not realistic to build AI agents and agentic management tools, such as agent meshes, from scratch. A faster, more straightforward path to agentic adoption is to work with external vendors that specialize in developing and supporting the necessary tools.

To this end, businesses should carefully assess the agentic vendor landscape and analyze the following key criteria:

• How mature the vendor's agentic technology is.
• How long the vendor has been in business and how much confidence its customers have that it will support its agentic tools over the long term.
• Which types of customer data the vendor or its software can access, and how it mitigates security and privacy risks linked to that data.
• Whether the vendor complies with key regulations that the business needs to meet.

8. Evolve your workforce management strategy

While it's too early to say exactly how AI agents will affect workforces, Goldman Sachs Research predicts that 6% to 7% of human jobs will become obsolete when businesses adopt AI agents to perform them.

At the same time, agentic technology will create a need for new types of human roles, such as those related to managing and securing AI agents. Hence, businesses must assess how AI agents will affect their workforces and evolve their business structures, hiring and worker retention strategies accordingly.

9. Optimize business processes

Most of the business processes that organizations currently have in place date back to the pre-agentic era. To maximize the value that AI agents create, businesses will need to restructure many of their processes.

Imagine that an AI agent can speed up one part of a workflow by 1,000%, but humans still manually perform other parts of the workflow that must occur in parallel. To create real value, the business would need to redesign the overall process by speeding up the parallel tasks or completing them at different times. Otherwise, the result could be a situation in which the agent-driven part of the workflow happens much faster while the overall workflow time remains the same.

Chris Tozzi is a freelance writer, research adviser, and professor of IT and society. He has previously worked as a journalist and Linux systems administrator.