As organizations move from generative AI to agents, success will depend on combining greater autonomy with stronger data context, governance, and oversight, says Elastic’s Ajay Nair
Since interest in personal AI agents grew earlier this year, the focus has shifted from what they can do for individuals to more serious enterprise questions: Can agents be trusted to act on behalf of businesses? Agent-building tools like OpenClaw show users how to build AI agents that perform tasks ranging from planning a day’s schedule to writing complex software code. But for organizations, the stakes are even higher.
Beyond personal use, organizations are now exploring ways to deploy agents throughout their business workflows to drive business outcomes, including increasing productivity, enhancing operations, and delivering measurable results.
At its core, an AI agent is a system designed to work. You can evaluate information, reason about tasks, plan next steps, and execute toward defined goals.
For organizations, this means moving from generative AI to agentic AI. It’s a shift from systems that generate or retrieve answers to systems that can take action in ways that impact customer experience, operations, and business outcomes, said Ajay Nair, general manager of Elasticsearch at Elastic.
Focus on impact, not activity
The appeal of AI agents is clear, but their usefulness will depend on where and how they are deployed. Instead of rushing to adopt the technology, Nair says there are three key factors that companies should consider before implementing agent AI.
First, organizations need to be clear about the value they want to create. Agent AI has different outcomes depending on where it is applied. For example, use it for product innovation to help teams move from idea to iteration quickly.
When applied to observability and security workloads, it has the potential to support stability in return by enabling organizations to manage risk, detect issues, and build resilience.
Next, companies need to identify workflows where agent AI can be applied effectively. This technology is ideal for processes with clear rules, repeatable steps, and well-documented playbooks. This is why many initial implementations started in engineering workflows such as software programming. Other use cases may require more foundational work before agent AI is ready.
Third, organizations must define how they will measure success. It is not enough for AI agents to produce more output or complete more tasks. For example, in customer support, agents may create hundreds of support articles, but if customer issues remain unresolved, the impact on the business will be limited.
Companies should not confuse activity with impact, Nile said. While moving quickly with AI may seem like progress, speed alone doesn’t translate to business value.
For many companies, excessive experimentation without a clear direction is costly and frustrating. Nair says this is often due to a lack of strategic thinking around AI adoption.
Therefore, your starting point should be concrete use cases and clear success criteria to ensure that your AI efforts are linked to business value rather than innovation for its own sake.
A practical approach to agent AI
AI innovation cycles are getting shorter, but the barriers to experimentation are getting lower.
For organizations, this means that agent AI cannot be treated as a one-time technology project. As technology develops and is applied to daily work, it must be understood, tested, and refined.
Ajay Nair, general manager of Elasticsearch at Elastic, suggests three priorities for enterprise leaders in this environment.
First, leaders need to understand how AI systems will work, including the data, context, and safeguards needed before implementation.
Second, you need to experiment in a contained space where you can leverage context to quickly achieve automation without exposing your entire organization to unnecessary risk.
Third, you need to stay curious. Rather than letting uncertainty about the technology delay decision-making, leaders should continue to learn how AI systems are evolving and where they can be applied responsibly.
While company leaders may want quick results from AI, a stronger measure of success is whether these efforts can create lasting business value.
As technology changes rapidly, Nile says leaders must remain curious about “how things work and how things change” to set their organizations up for long-term success.
Can AI agents be trusted?
Once a viable use case is identified, the next hurdle is trust. Can AI agents reach sound conclusions and act responsibly?
For organizations, trust is determined by the quality and relevance of the results that agents produce. It requires accuracy, transparency and, importantly, the right context, Nile explains.
In a business environment, the risks are much higher than in casual use. For example, errors in approving loans or processing employee repayments can have serious consequences. “This is different from casually using ChatGPT for holiday planning,” he added.
This concern is not just theoretical. Issues of accountability, control, and oversight become more pressing as AI agents are able to take action rather than just generate answers.
Countries around the world, including Singapore, are also paying close attention to their trust in AI.
The Department of Digital Development and Information launched the Model AI Governance Framework for Agent AI earlier this year, with the aim of guiding organizations to responsibly deploy agents. While the framework recommends technical and non-technical measures to reduce risk, it emphasizes that humans are ultimately responsible.
For enterprises, trust starts with the relevance of whether an AI application has the right context to properly perform its tasks. This means designing an AI stack that can leverage the right data, capture the right information, and generate the right insights for the purpose. As Nair says, it’s difficult to hire smart agents if the data behind them is poor.
Guardrails are equally important. These include controlling the types of data that agents can access, limiting read and write permissions for agents, and human oversight to ensure that proposed actions are safe, legitimate, and consistent with business rules and regulations.
“There are some tasks, such as coding, where you can trust AI more, but in general people need to know that the LLM they are working with is extracting good insights from the data they are being provided, and whether that data is relevant in the first place,” Nair adds.
Organizations therefore need to securely integrate AI solutions with their proprietary data, while ensuring that these systems capture relevant insights to solve specific business problems, he notes. “That’s the essence of context engineering.”
What is context engineering? Provide AI systems with the right information at the right time. For AI agents, this means having enough relevant, timely, and reliable context to make better decisions and take appropriate actions.
From context to accountability
The challenge is therefore not just to have more data, but to ensure that the right data is available to AI agents at the moment they need it.
Context engineering requires a strong data foundationHowever, organizations do not necessarily need to build all components from scratch. Even companies without mature data architectures can be running agent AI applications within months if they choose the right tools for their needs, Nair says.
The key is modularity. Modern AI tools enable organizations to assemble the necessary components for agent workflows such as data ingestion, retrieval, security, governance, and assessment.
For example, Elastic’s search capabilities are designed to understand the intent behind queries, retrieve relevant insights from stored data, and connect those insights to large-scale language models.
These features are integrated into Elastic’s Agent Builder and are intended to help organizations implement agent workflows faster while maintaining flexibility in the components and models used for specific tasks.
However, implementation is not the goal. AI agents must be continuously monitored and evaluated after they go live. Nair adds that each action or change made by an agent must be logged, checked for validity, and traced back to the data accessed.
This ensures security and Observability essential for agent workflows. With the right capabilities in place, organizations can monitor agent behavior, verify whether the agent was authorized to act, and ensure that agent behavior is appropriate, traceable, and aligned with business rules.
Learn more about how Elastic powers agent AI.
