Presented by Splunk
Every day, organizations learn things that their AI systems will never use.
A security analyst prepares investigations generated by AI. A network engineer identifies the source of recurring outages. The monitoring team finds that the pattern of latency, logs and infrastructure changes predict service degradation. The customer service team finds out which signs indicate that the increase is likely to increase.
Each minute contains important organizational information. But in most businesses, that information disappears into tickets, dashboards, chat threads, post-incident reviews and the minds of individual professionals. It may help solve an immediate problem, but rarely is it part of a reusable system that improves future AI-driven decisions.
That is the next challenge for the agent business.
The future won’t just be defined by who has the most talented model or the most independent agents. Many organizations will have access to similar parameter models. Many will deploy agents across security, IT, engineering, customer service, and business operations.
The real difference will be whether those agents can learn from the organization around them.
Not by constantly retraining the underlying model, but by capturing operational information, converting it to institutional information and making that information available for future agents, workflows, and decisions.
The agent business is not the only business that uses AI. It is a business learning about AI.
Agent entities allow AI systems to learn from them
The AI discussion is dominated by the model’s strengths: larger contextual windows, better reasoning, faster predictions, robust tooling, and more advanced agency behavior.
Those developments are important. But in business, the model is only one part of the system.
A model does not automatically know how a particular organization works. They naturally don’t know which corrective action solved last month’s end, which analyst fix improved threat investigation, which network signal preceded a service disruption, or which internal policy should override a potential recommendation.
That information is proprietary.
For agent systems to improve, organizations need a way to capture that information and make it reusable. In most cases, that doesn’t require changing the model itself. It requires changing the ecosystem around the model: the knowledge base, the retrieval layer, information, policies, monitoring lines, routing and workflows that change the way agents behave.
The model may remain the same. The surrounding learning program is smart.
Feedback loops turn every result into a teachable moment for agents
All agent workflows generate signals.
The agent receives the request. It finds context, reasons for possible actions, calls tools, and generates responses. The person accepts, rejects, or corrects that answer. Downstream systems report that the action worked.
All of that thread is important.
AI visualization gives organizations visibility into what happened: immediacy, response, mindset, tool calls, data sources, intermediate steps, failure modes and outcomes. Apart from that it seems, organizations cannot understand why the agent behaved the way it did, let alone improve it.
But being seen alone is not enough.
The biggest opportunity is to turn observed behavior into institutional knowledge. Tracing should not only help the engineer and operators to debug the agent. It should help the company to understand what the agent learned, what was corrected by the person, what was the result, and what should change before the next similar incident.
That’s a shift from monitoring AI to teaching AI.
In the agent entity, feedback loops connect action to outcome, outcome to information and information back to future action.
An effective learning program for all security, visibility and networking
Consider a service that experiences intermittent degradation.
The monitoring agent experiences unusual delays and error rates. A network agent identifies packet loss in every specific path. The security agent notices that at the same time the window includes suspicious authentication behavior and unusual traffic from a previously unseen source.
Individually, each agent has only a partial view. Together, they create a rich picture of performance.
If this incident occurs for the first time, human experts may need to intervene. A network engineer confirms that the packet loss was caused by a poorly configured routing switch. A security analyst determines that the suspicious traffic was not an attack, but the result of an internal service malfunctioning. SRE connects a network event to an application crash.
That decision contains information that the organization should not read again.
A mature agent learning system can capture traces, human configuration, topology context, security findings, visual signals and final corrective actions. It will maintain the relationship between those signals: delay pattern, network path, identity behavior, route change and resolution.
The next time the same pattern appears, the agents will not start from zero. They can retrieve a previous case, compare current conditions, recommend a proven diagnostic method and come up with a better context.
The lower bound model did not require retraining.
Business has learned.
An enterprise architecture for learning agents
A learning-oriented business needs more than a model or a chatbot. It requires an architecture that can capture experience, turn it into actionable information, connect that information to an operational context, and control how it changes future agent behavior.
Memory it keeps track of what happened: what the agent saw, what he did, where people intervened, and what consequences followed.
Knowledge bases turn that experience into reusable guidance, including playbooks, examples, policies, procedures, and evidence.
A data fabric connects the work area. Signal agents need live across logs, metrics, tracking, tickets, identity systems, security tools, network telemetry, collaboration platforms, and business applications. The data fabric makes those signals available, correlated, managed, and usable in context.
AI visualization describes how agents behave by capturing information, tool calls, intermediate steps, responses, feedback, and results. That seems to help organizations understand where agents succeed, where they fail, and what they need to improve.
I control plane it governs how learning becomes change: what information is developed, what incentives or policies are updated, what agents can use the new information, what permissions are required, and how changes are researched.
Together, these capabilities allow AI systems to evolve over time in a controlled, reliable way that allows the business to learn from its operations.
Organizations that learn quickly will win
The next era of AI will not be won by models alone. It will be won by organizations that can capture learnings across workflow, professional resolution, incident, investigation, and outcome.
More advanced businesses will not simply outsource more agents. They will build systems that allow all agents to benefit from the collective knowledge of the organization.
That means connecting operational data through the data fabric. It means looking at the agent’s behavior deeply enough to understand it. It means storing information in memory and entering it into knowledge bases. It means using a control plane to control how learning changes the agent’s behavior.
The future of AI is not a single autonomous agent that works alone. An ecosystem of agents, people, data and controls that learn over time.
Organizations that build that ecosystem will create AI systems that get better with every interaction. Not because the model is constantly changing, but because the business itself is getting smarter.
Read more about how Cisco Data Fabric is powered by the Splunk Platform speeds up the agent’s operations.
Hao Yang is Vice President of AI at Splunk, a Cisco Company.
Sponsored articles are content produced by a company that pays to post or has a business relationship with VentureBeat, and is always clearly marked. For more information, get in touch sales@venturebeat.com.
