Why context is critical to agentic AI – insights from Gartner

However, as the use of agentic AI increases, it brings a new challenge for Chief Data Officers (CDOs). Data now needs to be understandable, both semantically and in terms of context, in order to power AI agents to make better, faster decisions across workflows and processes, without human intervention. 

Context and semantics were one of the key topics at the recent Gartner Data & Analytics Summit 2026 London. This blog explores what they mean, how they impact AI success, and the role of the data product marketplace in delivering them successfully.

For many applications, data just needs to be accurate, reliable and high-quality. However, as data is shared across organizations and beyond the teams that first created it, new issues arise around semantics and context:

Semantics looks at the meaning of the data itself, rather than just its format or structure. It looks to answer questions such as:

• What does the data actually represent in a real-world context?
• How does it relate to other data?
• How should it be interpreted?

Semantics becomes important when data is shared outside specific teams or applications. For example, the term “customer” might mean one thing to sales, another to finance, and a third to marketing. All of this means that without agreed, shared semantics, every team operates from a different version of reality, leading to conflicting reports, failed analytics, and eroded trust in data. When it comes to AI, this could mean agents or models misinterpreting information or trying to compare disparate datasets without understanding their underlying differences. At best, this means inaccurate decision-making – at worst, hallucinations, potential bias and legal and financial damage.

Capabilities such as semantic layers, which translate technical raw data into business-friendly terms, and business glossaries, which standardize terms across the organization, all help support better semantic understanding of data.

Context goes beyond the meaning of data, to provide the information needed to interpret it correctly. It is broader than the semantic layer as it also covers governance, lineage, quality, and operational context. It is often stored as metadata, including factors such as who created the data, when, how often it is updated and what it is originally used for.

For humans, understanding the context of data is crucial to whether we trust and use it. For example, if a dataset is based on a survey of 90% of a company’s employees, it is clearly more trustworthy than one where just 10% have been polled.

Essentially, semantics is the literal meaning of words, phrases, and sentences in an organization, explaining what the language actually says. Context is the environment around it, such as the business situation, tone, or surrounding words, that shapes how we interpret that meaning.

Humans can often automatically understand the context of the data in front of them through their experience and knowledge. AI agents cannot. This means that failing to provide them with sufficient context leads to:

• Inefficiency in data management
• Greater financial costs
• Incomplete data governance
• Inaccurate, biased or hallucinatory decision-making, leading to legal and reputational damage

Given the interlinked nature of AI agents, and the lack of human involvement, this context needs to be present at every step of the agentic workflow to ensure accuracy and efficiency.

At the London Summit, Gartner released its prediction that by 2027, organizations that prioritize semantics in AI-ready data will increase their agentic AI accuracy by up to 80% and reduce costs by up to 60%.

In his presentation at Gartner London, analyst Andrés García-Rodeja outlined the three critical factors around agentic data – connection, context, and business value.

The Model Context Protocol (MCP) standard is currently used to feed AI agents with data. It acts as a transport mechanism that allows agents and Large Language Models (LLMs) to seamlessly connect to external systems. However, it does not have a built-in semantic layer, meaning that organizations need to incorporate these as separate parts of the data stack to help ensure data is understood correctly. In fact, Gartner believe that by 2028, 60% of agentic analytics projects relying solely on MCP will fail due to the lack of a consistent semantic layer.

Andrés García-Rodeja outlines that context engineering is the second critical step in AI agent-ready data. Data must be enriched with contextual information so that AI agents can easily understand and choose the best assets for their needs. For example, if there are two, similar data assets, which is the most used and recommended by the business? These contextual signals help guide agents when searching for information. Factors such as governance, provenance, usage, and behavior all fit into this context layer.

Currently, just 22% of organizations feel that generative AI provides them with significant value. Overcoming this gap is therefore vital to scaling AI and achieving its full impact. As well as adding semantics and context, deploying knowledge engineering technologies such as knowledge graphs and ontologies help increase trust and understanding of data, driving value.

Data product marketplaces centralize data assets, including data products, and make them easily available to both AI and humans, through an intuitive, e-commerce style interface and machine readable information. 

Data product marketplaces, such as Huwise’s solution, are essential to delivering context and semantics in eight different ways:

• Data products: Marketplaces provide seamless access to data products, high-value data assets, that have contextual information built-in, a specified data owner, and a clear data contract around usage and quality
• Usage intelligence and lineage: By monitoring which data products receive the highest usage, highlighting use cases, and tracking lineage, marketplaces deliver usage intelligence. They provide understanding of which data is used, by whom, and in which context, helping AI agents prioritize the best data for their needs. Data teams can also see where more data assets are required and which need to be updated to include better context and to build trust
• Business glossary: Marketplaces have a built-in business glossary, providing a standardized vocabulary around data terms and what they mean in that organization’s specific business context
• Semantic layer: Marketplaces retrieve and apply semantic layers defined in other tools, ensuring consistency across the data stack
• Metadata management: Marketplaces include powerful metadata management functionality, including support for common international standards
• Feedback: Marketplaces enable users to rate data and give feedback, all providing information that helps support the context layer 
• MCP server: Data can be easily consumed by agents and LLMs through a MCP server, providing machine-readable access to data
• Data context: Marketplaces automatically recognize data types, and use this to deliver context to results. For example, if data contains zip codes it can be used to localize results, bringing ontology into the context layer.

As part of its strategy to underpin context and semantics, Huwise is an Atlan Context Layer Partner. This means its data product marketplace delivers business usage information and context to ensure AI agents are guided to the best data. The combination of Atlan’s Enterprise Context Layer – unifying lineage, business definitions, quality signals, and governance rules into a single graph – with Huwise’s usage intelligence creates a complete picture: AI agents know not only what data means, but which data has been validated by the business in practice.

Agentic AI offers the ability for organizations to accelerate decision-making, increase efficiency and deliver better and more effective service to customers. However, without a focus on the context and semantic layers, even the most accurate data risks being misunderstood or misused by AI. Building a robust, end-to-end context layer is therefore vital to agentic AI success and achieving business ROI.