Agentic AIs, a form of technology designed to run specific functions within an organization without human intervention, are gaining traction as enterprises look to automate business workflows, augment the output of human workers, and derive value from generative AI.
Analyst firm Forrester named AI agents as one of its top 10 emerging technologies this year, but it has a warning for companies focused on adopting them: Don’t go it alone. Three-quarters of the organizations that try to build AI agents in house will fail, according to Forrester’s 2025 predictions for AI.
Companies that fail to build their own AI agents will turn to outside AI consulting firms to build custom agents for them, or they will use agents embedded in software from their current vendors, write Forrester analysts Jayesh Chaurasia and Sudha Maheshwari.
“Savvy firms will grasp current limitations and lean on their vendor and systems integrator partners to build agents at the cutting edge of this technology,” they write.
Building AI agents is a complex process, and many organizations don’t have the AI expertise in house to finish the job, they add.
“Agentic AI is all the rage as companies push gen AI beyond basic tasks into more complex actions,” Chaurasia and Maheshwari say. “The challenge is that these architectures are convoluted, requiring multiple models, advanced RAG [retrieval augmented generation] stacks, advanced data architectures, and specialized expertise.”
In addition, the power of agentic AIs is still in its infancy, they say. It may be another two years before they have “any chance of meeting inflated automation hopes,” Chaurasia and Maheshwari write.
The value of DIY agentic AI
Still, some companies see opportunity in the build-it-yourself approach. Goldcast, a software developer focused on video marketing, has experimented with a dozen open-source AI models to assist with various tasks, says Lauren Creedon, head of product at the company.
For example, Goldcast uses one AI model to transcribe videos, another to write a blog post based on a video, a third to create social media posts, and a fourth to identify the people in the video through facial recognition, she says. The goal at Goldcast is to link all these AI models and turn them into agents that do their assigned tasks without human prompts, she says.
Goldcast has taken the abilities of each of these AI models and used specific features for its own use cases and workflows. The company isn’t building its own discrete AI models but is instead harnessing the power of these open-source AIs.
Building on open-source models is a more efficient way to tap into the power of agentic AIs than creating AI agents from scratch, Creedon says.
“I don’t want people to think of [AI] as hard and a specialized thing that only people with PhDs can work with,” she says. “The more people who are enabled on how to work with it, and the more teams that work with it, the better outcomes will get, not only for business operations, but for customers.”
But Creedon agrees with Forrester’s assessment that building AI agents can be a complex process. Organizations will need a fully formed MLOps plan, and some companies may not have the expertise to do it themselves, she says.
Advanced teams will be required to “take a number of these different open-source models and pair them together in a workflow,” Creedon adds. In many cases, organizations will need to turn to outside specialists to set up AI agents.
Still, it’s possible to do it yourself, says Senthil Kumar, CTO and head of AI at Slate Technologies, a data analytics provider for construction and related industries.
Slate Technologies began rolling out its own AI agents three years ago, even before the AI boom kicked off with the release of ChatGPT.
“What has been an aspirational technology to look forward to a few years ago is being realized,” Kumar says.
With several LLM AIs now available, smart companies can experiment with them and train autonomous agents based on their specific needs, he says.
“We are fortunate to be able to stand on the shoulders of giants and learn from others’ experiences in the space.” Kumar adds. “Start with one [AI model], and you can start tailoring its behavior. You know your ecosystem much better than a generic solution that exists outside, with consultants outside.”
Humans in the loop
One key to building successful AI agents is human supervision, which is always necessary even when agents are built to run autonomously, Kumar says. Organizations can’t build an AI agent, unleash it, and forget about it; instead, they need to check the results and continuously find ways to improve them.
“It’s a collaborative process of evolving between the whole AI ecosystem and the human counterparts,” he says. “The focus would be on how those agents would learn, the knowledge acquisition of agents, and how the agents are going to be able to disseminate knowledge.”
For many companies, however, the decision to build their own AI agents or work with a consultant isn’t an easy one, says Chris Ackerson, head of AI at AlphaSense, an AI-powered market intelligence firm.
Large companies may be tempted to roll their own highly customized agents, he says, but they can get tripped up by fragmented internal data, by underestimating the resources needed, and by lacking in-house expertise.
“While some companies may achieve success, it’s common for these projects to spiral out of control in terms of cost and complexity,” Ackerson says. “In many cases, buying a solution from a trusted partner can help organizations avoid the pitfalls of builder’s remorse and accelerate their path to success.”
AlphaSense has trained its own AI agents, but many companies lack internal expertise, he says. In addition, organizations may project the development costs but ignore the cost of ongoing maintenance, he adds.
“This is the largest cost, as maintaining AI systems over time can be complex and resource-intensive, requiring constant updates, monitoring, and optimization to ensure long-term functionality,” Ackerson says.
Partnering with an AI provider can give companies access to proven, ready-made agents that have been tested and refined by thousands of users, he contends.
“It’s faster to implement, less resource-intensive, and comes with the added benefit of ongoing updates and support — freeing companies to focus on other critical areas of their business,” he says.
The value of a partner
Many organizations won’t need to train their own AI agents, says Adnan Masood, chief AI architect at UST, a digital transformation provider.
“Reinventing the wheel is indeed a bad idea when it comes to complex systems like agentic AI architectures,” he says. “These architectures are inherently intricate, involving a multitude of components.”
One of the many challenges, he says, is to implement robust memory management in an agentic AI system. The process goes beyond storing and retrieving information and includes intelligently managing context, understanding the relevance of past interactions, and dynamically adapting the AI’s responses based on an evolving knowledge base.
In addition, building an agentic AI from the ground up would involve designing complex data structures, implementing efficient search algorithms, and fine-tuning the AI’s ability to interpret and prioritize information, he adds. This would require organizations to have specialized expertise in machine learning, natural language processing, and data engineering.
“By turning to specialists or adopting pre-built solutions, or tapping into the open-source ecosystem, they can leverage the expertise and experience of those who have already navigated these challenges, ultimately increasing their chances of success,” Masood says.