Amid all the excitement and debate, one thing is clear: Agentic AI is set to be yet another extraordinary step forward for artificial intelligence (AI). By enabling organizations to automate key business processes fully, it will unlock new opportunities to boost productivity and efficiency, bolster cybersecurity, and supercharge growth.
The precise impact of this new generation of AI agents will vary from business to business. In some cases, it will mean building specialized agents for specific tasks within larger processes, such as making a journal entry, drafting an RFP, or responding to customer inquiries.
In others, it may involve deploying fully autonomous agents capable of orchestrating an entire process from end to end. For example, in the areas of compliance, agents could identify data needed to review compliance against a set of standards, identify gaps, report on them, and remediate issues without any required human involvement to complete the task.
Yet, what’s true across the board is that these agentic systems will be transformative, freeing up workers from repetitive tasks, rapidly assessing market risks and opportunities, and giving leaders connected and real-time data to sharpen their decision-making. And they can do it all with limited or no direct human intervention.
Subscribe to the Daily newsletter.Fast Company’s trending stories delivered to you every day
Privacy Policy
|
Fast Company Newsletters
START TO END GAME
Successfully integrating agentic AI into operations requires a fundamental shift in organizations’ overall strategies and mindsets. Whereas with generative AI (GenAI) tools, many leaders have focused on task augmentation within areas like customer service and product development, agentic AI gives them the chance to be far more revolutionary. Rather than looking at a process and asking, “Where can I use AI to improve it?” they can achieve an even greater impact by completely redesigning the process using agentic AI from start to finish.
Customer churn analysis and prevention is a great example—not the least because it’s an issue that almost every organization can relate to. Traditionally, this is a process requiring multiple teams across multiple business areas, including data engineers, data scientists, marketing analysts, and executives. It also involves various steps—from data exercises like identifying at-risk customers and analyzing any trends to business and marketing programs like designing intervention strategies, executing targeted campaigns, and monitoring effectiveness.
With agentic AI, the whole process could be executed by just five agents, with each responsible for a different step and all working autonomously yet cohesively together. This not only improves efficiency but also ensures real-time, data-driven decision-making.
HUMANS ABOVE THE LOOP
But if all that sounds like a no-brainer for rapid adoption, it’s not—at least not yet. Certain barriers must be overcome and governance is at the top of the list.
Research has shown that training a large language model (LLM) to do one malicious task can lead to it becoming an entirely malicious AI, with far-reaching consequences for a firm’s stakeholder relationships, security, and more. Now imagine the same thing happening in an agentic system involving multiple AI agents collaborating and orchestrating with each other.
Establishing a robust (not to mention constantly evolving) framework for responsible deployment will therefore be vital for organizations when it comes to creating trust and implementation at scale. Likewise, transparent and rigorous systems for human supervision oversight are needed to ensure the safety of data inputs being used, along with the outcomes being delivered.
advertisement
It’s an environment built on having humans above the loop rather than humans in the loop.
GARBAGE IN … YOU KNOW THE REST
The expanding global regulatory landscape may have a stymieing effect too, with certain geographic areas facing greater constraints on how deeply they can integrate agentic AI. Look no further than the EU’s Artificial Intelligence Act. While the verdict is still out on exactly how impactful this will be to the region, there are already concerns about how this regulation will increase costs and slow innovation speed.
And, of course, there remains the pressing issue of data. We’ve all heard the refrain “garbage in equals garbage out” many times before when it comes to AI—and it’s no different with agentic AI. Indeed, whether it’s for specific process agents or fully autonomous orchestrators, agentic systems rely on having high-quality insights to reason over.
Organizations need to invest more than they have in the past in the quality, access, and the data they are leverage for agentic AI systems. Further, new approaches are necessary to transform their data into actionable knowledge. Caring and modernizing today will create more effective AI agents with less risk using simple steps for building this solid data foundation.
READY AND WILLING TO BE ABLE
Highlighting these barriers is by no means an attempt to pour cold water on the flames of innovation. The potential that agentic AI offers firms to enhance performance and drive competitiveness is undeniably exciting. Yet, we must also be realistic in recognizing there is work still to do if we are to seize that opportunity.
Rather than aspiring for full adoption, leaders may benefit more by viewing the next several months as a time for assembling agentic knowledge and preparing for readiness. First, by investing in building the necessary governance frameworks, data infrastructure, and process engineering to support agentic systems. Second, by working with policymakers at the state, national, and international levels to overcome the challenges of a diverse global regulatory landscape.
As for where we started, namely the much-asked question of whether 2025 is the year of AI agents, the answer right now is that this is the time to enable the ability to scale AI agents. The agentic age is approaching fast and will likely reshape business operations forever. Firms that get ready for the revolution now will be best able to lead it in the future.
Traci Gusher is the Americas AI and Data Leader at EY.