Context engineering: The real reason your AI agent delivers or disappoints

AS WE enter 2026, the corporate landscape has shifted from experimenting with generative artificial intelligence (AI) to deploying autonomous AI agents capable of performing complex tasks for employees and customers alike. This is a frontier where there is no margin for error – and where the path to value is easily obscured by hype.

For organisations to unlock real returns and avoid wasted investment, the journey must begin not with the tool, but with a clearly defined problem.

Integrating AI, like any technology, means changes to existing processes and legacy systems. Without buy-in from the people it is supposed to help, it is set up to fail. The complex operational processes that businesses have built over time must also evolve to accommodate AI – and deliver value through it.

Ravi Rajendran, area vice president, ASEAN, Hong Kong, Taiwan, and Korea at Elastic shares his thoughts on how organisations can implement AI agents successfully.

Start with the outcome in mind

Rajendran advocates a “work backwards” philosophy. He explains: “If an organisation has a problem in mind for an AI agent to solve, they need to work backwards to find out what it needs to succeed.

“Keep the intended outcome and the purpose in mind, then pull together the pieces for an effective solution, like methods to retrieve context and choosing a large language model (LLM) to help an agent achieve its intended purpose.”

This problem-first approach ensures AI efforts are not only innovative but meaningful. Businesses should anchor their AI strategy to known, concrete challenges – improving healthcare diagnostics, optimising energy consumption or enhancing accessibility, for example. 

Once the outcome is defined, the organisation can assess what is needed to get there.

A solid foundation with contextual data

The industry has learned that an LLM’s general knowledge is rarely sufficient for enterprise needs. An LLM can write a poem, but it cannot inherently know a client’s portfolio or a company’s proprietary security logs. 

This is where context engineering becomes the critical differentiator between an AI that hallucinates and one that delivers.

“The accuracy of AI is fully dependent on the context you provide it,” says Rajendran. “Without the right context, AI models struggle with relevance, often producing inaccurate responses. 

“We need to engineer for context, providing the right information at the right time and in the right format to empower an AI agent to perform complex, use case-specific tasks.”

Context engineering is the practice of giving AI systems exactly the right information – documents, databases, emails, code – at the right moment. The process involves intelligently selecting, organising and delivering what the AI needs to make sound decisions, without overwhelming it with noise. 

A security-focused agent, for instance, needs access to threat intelligence, security logs and reports to identify the alerts that genuinely require attention.

Done well, context engineering is the difference between an AI that offers generic responses and one that delivers accurate, actionable answers grounded in an organisation’s own data.

Tackle complexity by building simple

AI agents demand far more precision and context than general-purpose LLM applications – but that does not mean they need to be difficult to build.

Consider a financial manager preparing for a client meeting. A custom-built AI agent can process a query about market movements affecting a client’s holdings, retrieve relevant contextual data from their portfolio, and use an LLM to generate a clear, natural-language overview of the key trends.

Rather than building every function from scratch, companies like Elastic offer agent builders with pre-assembled components that help organisations create business-ready agents quickly. 

Elastic’s Agent Builder lets teams assemble capabilities – natural language processing, data retrieval across disparate sources, information processing and reasoning, and a choice of LLM – much like putting together building blocks.

Underpinning this is Elastic’s developer-centric, open-source Elasticsearch platform, which helps organisations move from pilot projects to scaled deployments.

Auditors offer a compelling use case. Once required to manually sift through vast amounts of data to track changes in contracts or pricing over time, they can now use search and AI to scan information, surface patterns and generate concise overviews. 

The auditors focus on reviewing and verifying the specific data points the system flags – a “human-in-the-loop” model that augments individual capability rather than bypassing it.

“Beyond the applications of search and AI, we’re always pleasantly surprised to see how our customers and partners are building with our technology and the outcomes they achieve,” says Rajendran. 

“From enhancing in-app search capabilities with the ability to understand meaning and nuance, to improving knowledge management, observability and security through autonomous AI agents, our search and AI capabilities are proving especially valuable to retrieve highly specific, contextual data to identify trends and insights that might otherwise be missed.”

The future of agentic AI

The move toward agentic AI is as significant a transformation as the initial shift to the cloud. As these agents mature, they will do more than automate manual tasks – they will help organisations uncover untapped opportunities and new revenue streams.

“The sooner organisations decide on the outcomes they want, the faster they can leverage existing tools and resources to accelerate their AI journey,” says Rajendran. 

“By focusing on a solid data foundation and keeping the ‘why’ at the centre of every project, organisations can ensure their AI investment leads to lasting transformation rather than passing hype.”

Find out more about Elastic’s search and AI capabilities at Elastic{ON} Singapore 2026 on March 17. Entry is free!