One exciting and fast-emerging technology stands out as business leaders look for new ways to drive innovation: AI agents.
For the past two years, technology teams have had their hands full with early-stage AI projects and, in particular, generative AI. Now, a growing number are evaluating AI agents, the beginning of an agentic wave of innovation and efficiency that could go on for years.
It’s conceivable that AI agents will eventually have a place in almost every business—not unlike Bill Gates’ original vision of putting a PC on every desk and in every home. Now, as then, the potential benefits include productivity improvements for both individuals and entire companies.
Various pieces have to come together to make this happen. We can expect to see progress in three areas: GPUs and their insatiable energy requirements; the learning algorithms used to train agents; and domain-specific data that make agents suitable for a particular industry or use case.
All of this will be driven by a broad ecosystem, from AI startups such as OpenAI and Anthropic to hyperscale cloud providers and many others in between. Companies that specialize in data management and data professionals will play a vital role because agents are only as good as the data that drives their actions.
Learning to train and refine agent behavior
Let’s take a closer look at three pillars of the agentic future: GPUs, algorithms, and domain-specific data. The activity is picking up in each of these areas.
We’re witnessing the build-out of data centers and the energy grid to support growing demand for the GPUs that power AI development and training. In fact, the energy requirements are so enormous that hyperscalers have begun looking at nuclear energy as a potential source.
No doubt, we will continue to see technical innovation in hardware architecture to enable models at a range of scales, from even bigger large language models (LLMs) to smaller, local, and private models.
At the same time, algorithms will flourish in the agentic world. Today, the primary learning algorithms include transformers, which convert natural-language inputs into outputs, and deep learning, the application of neural networking to big data.In this realm, reinforcement learning will play a growing role in how agents learn to make decisions and take actions.
That leads to domain-specific data, which will become increasingly important as agents take on different roles and tasks. For example, a human resources agent must know the skills and experiences that are essential for job success, while a product agent must be versed in everything from installation to maintenance. Businesses will need to develop expertise in training and refining agent behaviors with this kind of domain-specific data.
How will they do that? Development teams will have to establish and refine processes as they advance from prototype agents to more sophisticated ones. The workflow might include simulations based on both real-world data and computer-generated synthetic data. All of that data must be tested, annotated by humans, and then fed back to the agent for reinforcement learning.
The winners in agent strategy and deployment will be organizations that understand this kind of data-refinement loop and, importantly, the human experience of interacting with agents.
Improvement loops and simulations
It’s a challenge to provide the detailed context needed by agents to excel at their tasks. A contact center agent may need access to call logs, customer profiles from the CRM system, purchase records, and previous tech support inquiries—all aggregated in near-real-time to enable an excellent customer experience.
Yet, this context-setting is also an opportunity, because those that are adept at it will have an advantage. Organizations can get a head start by ensuring their data foundations are strong and implement best practices. To support this requirement, we will see the evolution of cloud-based tools for orchestrating agent improvement loops and running simulations with structured and unstructured data, AI models, and context stores.
Another data-dependent factor is collaboration—both agent-to-agent and human-to-agent. In the contact center example, this might entail a customer-support agent interacting with a business-department agent. The data must be logged on both ends and potentially analyzed by yet another AI agent for continuous improvement. The key to this kind of automation is enabling agents to talk to each other in the same “language.” This is where data exchange formats and protocols come in along with, not to be forgotten, privacy controls.
Trusted AI agents need trusted data
None of these developments will come to pass if AI agents aren’t trusted by employees and customers. For this reason, responsible AI powered by high-quality data must be a priority, now and for as far as we can see into the future.
Don’t be surprised if there are miscues along the way. Potential agent gotchas include use of copyrighted data in AI models, bias, and hallucinations. So, project leaders must be attentive to safeguards, governance, regulations and best practices in data management.
Success with AI agents will depend on having the right data in the right place at the right time. Of course, that’s long been a strategic goal for CIOs and CTOs. Now they have an even better reason to make it happen.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
Read more:
- IBM CEO: DeepSeek proved us right—AI is not about big, proprietary systems
- Honeywell CEO: AI will transform industry at scale beginning in 2025
- AI ‘prompt and pray’ hasn’t cut it in the enterprise, but we’ve found the missing puzzle piece. Mass deployment is next
- CEOs: Implementing AI without understanding this one thing could cost you
