Agentic AI uses sophisticated reasoning and iterative planning to autonomously solve complex, multi-step problems. By absorbing massive amounts of data from multiple sources, the technology can build strategies, analyze challenges, and execute tasks in an almost endless range of business and research sectors, including supply chains, cybersecurity, and healthcare.
Traditional AI systems typically excel at narrowly defined tasks under tightly controlled conditions, says Michael Craig, staff scientist at AI drug discovery firm Valence Labs. Agentic AI systems aren’t restricted to a single, narrow purpose. “They can identify which questions to explore, what experiments to run, and how to adjust a methodology as new data emerges,” he notes via email.
Agentic AI functions like a workflow compared to other AI applications, says Joe Fernandes, vice president and general manager at enterprise open-source software provider Red Hat’s AI unit. “Rather than a typical generative AI model generating a single response to a question, an agentic AI system may execute several steps on its own to complete the task,” he explains in an email interview. This could include analyzing the request, mapping out a strategy, and executing the task, which in itself could be calling out to additional models or external systems, such as a search engine or querying a database.
A Force Multiplier
When fully realized, agentic AI can be a force multiplier to an extreme degree, Fernandes says. “Looking at it from the perspective of a traditional enterprise IT organization, it’s like having an incredibly specialized individual — or team of individuals — that doesn’t mind having the same task, every day, with no creativity or scope expansion.”
Given its powerful and wide-ranging abilities, agentic AI presents an opportunity to advance scientific research by analyzing petabytes of data, formulating hypotheses, and pinpointing salient patterns in an asynchronous manner. “This has the potential to accelerate advancement in data-heavy fields like biology, chemistry, and drug discovery,” Craig says. “Furthermore, agentic AI can update plans based on intermediate findings without needing continuous human supervision, which can result in a broader exploration of possible solutions.” Perhaps most importantly, by testing ideas in simulated environments, agentic AI can lower reliance on expensive wet lab experiments, improving the likelihood that subsequent experiments will drive insight.
Agentic AI can also free IT team members from maintenance and other low-level tasks, Fernandes says. Instead, staff can work on integrating new systems or applications, engaging more closely with customers, and handling other important duties. “In this scenario, agentic AI takes on the unpleasant tasks of IT work and lets a technology organization drive incredible value for the broader business rather than being stuck in cycles of system maintenance.”
Over time, agentic AI has the ability to improve its performance by learning from experience, becoming increasingly effective at achieving desired outcomes, says Marinela Profi, global AI market strategy lead at business analytics software and services provider SAS in an online interview. “For example, it might reschedule deliveries to avoid traffic or change a factory’s production plan if demand rises.”
First Adopters
Initial agentic Ai adopters will likely be enterprises looking to maximize their AI investments, boost productivity, and tackle complex business challenges, predicts Lan Guan, chief AI officer at business advisory firm Accenture. “These organizations are particularly interested in solutions that can scale across multiple functions and operate with minimal human oversight,” she notes via email.
Enterprises across a wide range of verticals are most likely to be the first to commit to agentic AI, eying the potential for reduced costs, Fernandes says. “Looking at agentic AI in the long term, it’s feasible that almost every organization in nearly every industry can benefit from adopting agentic AI agents in some fashion.”
First Steps
The best way to get started with agentic AI is by establishing a strong foundational infrastructure and resilient data management practices, Guan says. “Organizations are at varying stages of readiness, and those with a robust enterprise platform architecture are better positioned to ensure seamless accessibility to foundation models.”
An easier approach to agentic AI is simply experimenting with the technology. “The good news here is that much of the innovation surrounding agentic AI, and AI in general, is happening in open source,” Fernandes observes. He points to several emerging agent tools/ frameworks, including CrewAI and LangChain, among many others.
On the downside, agentic AI faces some of the same challenges as other generative AI use cases. “The underlying GenAI models need to be trained and tuned on your data and deployed for inference across a hybrid environment that may extend from public clouds, to private data centers and out to the edge,” Fernandes explains. “This needs to be done in a cost-effective way to ensure a positive ROI, which is a challenge given that this generally requires accelerated compute hardware, namely GPUs.”
Adopting organizations also must possess the internal skills and resources needed to effectively train models on their data, as well as a clear deployment strategy. “AI agents offer a solid pathway to production AI, but the constantly evolving market, from new model introductions and technologies to training and RAG-type techniques, means that most IT organizations are currently being very deliberate in their pursuit,” Fernandes says.