Many organizations will look to develop a select set of strategic AI Agents. My question for you is, are you ready to lead this effort and deliver AI experiences to customers and employees? Additionally, will your organization avoid last-generation mistakes when developing proprietary technology capabilities?

Organizations still look at technology investments and transformation initiatives through a build versus buy lens. But when it comes to AI agents, the approach is neither, and both – because developing and supporting AI agents and progressing toward more autonomous agentic AI requires a hybrid approach.

I review this approach in the white paper, Make AI Work: Unified Search & Retrieval for the Enterprise. The paper covers how top-performing companies compete on improving end-user satisfaction and customer loyalty by centralizing real-time access to information for AI agents, ML models, and people.

Simplifying how digital trailblazers lead AI agent development

AI agents are services designed for specific roles and job functions. They accept and respond with natural language and connect to APIs to perform actions. Developing AI agents has four steps.

1. Define the role you want the AI agent to play, its responsibilities, the governance on what types of decisions it can make on its own or with a human in the middle, and the validation rules to test for prudent decision-making.
2. Unify search and retrieval of information that the AI agent uses to reason its way through a task, as well as for processing natural language inputs and responding with its recommendations.
3. Connect the AI agent to APIs and automations to perform actions.
4. Develop a continuous testing and monitoring strategy that validates recommendations, shares performance metrics, and indicates when an AI agent needs to be upgraded.

Many organizations will use commodity AI agents from SaaS platforms to support HR, finance, marketing, and other departmental workflows. But even these AI agents will require role alignment and access to more information than what’s stored in the platform. In other words, configuring and improving the performance of these agents will benefit when organizations integrate them with a unified search and retrieval platform.  

Avoiding mistakes when developing proprietary AI agents

Proprietary AI agents to support customer and proprietary workflows will require implementing all four steps. And that’s where some organizations will make mistakes, electing to customize the full stack of the AI agent instead of looking for onramps and accelerators. In fact, Gartner predicts that by 2028, more than 50% of enterprises that have built their own language models from scratch will abandon their efforts due to costs, complexity, and technical debt.

While public clouds offer the platforms for data pipelines, data management, and access to LLMs, there’s quite a bit of plumbing to get it configured and optimized. Most organizations don’t have the technology skills and patience to custom-engineer the foundations, and many don’t require the scale to dive into these complexities.

Worse, AI agents are not one-time development efforts. The models require updating with the most relevant information and using feedback to optimize their decision-making. Develop the plumbing, and expect to upgrade it and the slew of underlying technology platforms regularly.

Avoid implementation complexities when developing AI Agents

Instead of building the data and AI model’s plumbing, StarCIO recommends buying and configuring the technology that makes relevance possible.

When evaluating platforms, look for out-of-the-box integrations to cloud and SaaS providers, such as AWS, Microsoft Dynamics 365, Salesforce, SAP, ServiceNow, and Shopify. Then, consider the APIs that AI agents will need to process natural language requests and responses, including a search API, a passage retrieval API, and an answer API.    

Now, instead of investing all the engineering effort around the plumbing, the development team can focus on the AI agent automation capabilities and decision-making quality.

• For customer-facing AI agents, the development team should concentrate on optimizing journeys, improving experiences, and exceeding customer satisfaction.
• When developing strategic employee-facing AI agents, the team must target KPIs on delivering business value, time-to-decision, decision quality, change management effectiveness, and employee satisfaction.

In the white paper, I share example financial metrics and business outcomes in four categories: revenue and growth, customer happiness and support, employee satisfaction, and IT innovation and efficiency. The paper also includes five key steps for agile business, data, and IT teams to collaborate on when developing unified search and retrieval experiences.

AI agents are not just for tech companies; I expect leading companies in healthcare, financial services, telecommunications, retail, and manufacturing to develop both customer- and employee-facing AI agents. These organizations will focus on the roles, experiences, and data, while selecting accelerating platforms to manage the AI and integration complexities.

This post is brought to you by Coveo.

The views and opinions expressed herein are those of the author and do not necessarily represent the views and opinions of Coveo.