AI agents are the future of enterprise software. Spending on AI initiatives accounts for the second largest proportion of IT spending, behind only security, and it is growing faster than any other IT segment according to ISG’s 2026 IT Budgets and Spending study. ISG asserts that through 2028, almost all software providers will add agentic capabilities to enable automation and streamline operations. Software developers are implementing AI agents to accelerate the delivery of business outcomes, improve operational efficiency and reduce costs associated with tasks that would otherwise require manual intervention.
The evolution of AI from deterministic to generative/probabilistic to agentic has also transformed how enterprises need to handle data. Large volumes of data are required to train models, while data freshness is important to inferencing in interactive applications, and data quality is fundamental to ensuring that the output of agentic and generative AI initiatives can be relied upon.
AI agents may require governed access to trusted enterprise data and the context necessary to understand and appropriately use data. However, integrating and utilizing decades of accumulated knowledge and data presents a significant challenge due to the complexity of legacy systems and existing business practices. This is driving the need for intelligent data software that is agile, adaptive and automated, forming an integral part of agentic processes.
ISG Research defines AI agents as software entities that perceive their environment, make context-aware decisions and take automated actions with indefined business and policy constraints based on perception and reasoning. While many existing AI applications rely on simple rule-based logic and are primarily reactive, AI agents complete the last mile of the sense-analyze-decide-act system paradigm to automate the execution of complex business processes and the delivery of business outcomes.
To do so effectively, AI agents require knowledge of trusted business processes, objectives, terminology and historical decisions. An enterprise knowledge layer combining metadata, data governance and data product capabilities to understand context intent as well as regulatory and policy-based constraints to close the gap between what is technically possible with AIagents and the proven delivery of business value.
Organizations are primarily leveraging AI to achieve efficiency gains and cost savings, automating repetitive tasks and reducing manual labor while enabling staff to focus on higher-value tasks and functions. To successfully automate the delivery of business outcomes, AI agents must deliver proven tangible business value that can be measured in relation to clearly defined business goals.
Participants in ISG’s 2025 State of AI Study identified three corelessons learned from AI initiatives that they would pass on to their peers: the need to define clear goals and key
Treating data as a business discipline and maintaining focus on a clearly defined business problem is essential. Traditional approaches to data and AI initiatives have invariably been data centric: data teams tend to focus on cataloging and curating data, standing up data platforms and building dashboards. Too often, this overemphasis on assembling and integrating data can lead to the original business requirement falling by the wayside.
The concept of data products facilitates an outcome-led focus for data and AI initiatives. Applying product thinking to data initiatives establishes the desired business outcome as the focal point for the assessment of the processes and data required to deliver that outcome, as well as identification of the organizational goals and metrics that will be used tomeasure its success. Although data teams remain responsible for implementationof data products, they do so in the context of, and alignment with, a strategicdirection defined and led by business leaders and a focus on the processes,outcomes and KPIs aimed at fulfilling the business requirement.
A metadata-driven data knowledge layer that automates data governance can help deliver a holistic business-level view of data production and consumption to support the identification and measurement of metrics and KPIs for data and AI initiatives. This holistic view enables data administrators to understand the use of data in data products and AI initiatives and provides business leaders with agreed-upon and trusted metrics with which they can identify and measure the value generated by data productsand ensure that projects are delivering the desired business outcomes.
Ensuring that enterprise data can be used for AI initiatives is the most significant current data challenge according to participants in ISG’s 2025Market Lens Data and AI Program Study. Too many enterprises struggle with data that is fragmented, inconsistent and not easily accessible. To be AI-ready, data needs to be clean, well-organized and compliant with regulatory standards. Additionally, it needs to be aligned to the specific requirements of the use case and its business objective.
The metadata management, data lineage, data governance, data quality and usage intelligence capabilities provided by a data knowledge layer are essential to creating a holistic understanding of data production and consumption to support the implementation and measurement of AI agents. Metadata-driven data inventory is a core capability that surfaces information from underlying data stores to enable search-based discovery and classification. While metadata-driven datainventory capabilities were initially focused on facilitating search-based access to data by business analysts and decision-makers, they are also proving essential in providing AI agents with automated access to trusted enterprise data and information.
Identifying, classifying and maintaining a map of relationships between data assets, business processes and decision-making tasks is also essential to enabling AI agents to understand business context and is driving a focus on knowledge graph capabilities that facilitate automated data discovery and enhance intelligent operations.
Additionally, data management tasks are prime candidates for automation and acceleration using AI. Data management tasks are often repeatable and routine but can also be mundane and time-consuming, especially at scale. Examples include data profiling and tagging; the identification of personally identifiable information; the creation and implementation of data quality rules; and the documentation of data integration pipelines. Automating data management tasks with AI can improve productivity for data teams by automating complex but essential tasks and can enable data experts to focus on higher-value tasks that more directly support business goals and innovation.
Data agents are increasingly being adopted to bring the advantages of agentic AI to data management processes. For example, data documentation agents can automate the classification of data with agreed business terms and descriptions, accelerating the conversion of raw data assets into trusted AI-ready data. ISG asserts that through 2027, enterprises will prioritize data intelligence software providers offering agents to automate the discovery and classification of data assets.
Providers that fail to address data quality may also be at a disadvantage. Data agents can be used to orchestrate data quality assessment, identification and remediation by operationalizing and automating the application of data quality rules and policies. Data agents can also automate the processes involved in packaging, documenting and preparing data assets to be shared and consumed as trusted and certified data products, in accordance with pre-defined business outcomes.
As enterprises accelerate their use of AI agents, a data knowledge layer that can support the creation of AI agents while also being part of the agentic process will quickly become not just desirable, but essential. ISG recommends that all enterprises exploring the potential benefits of AI agents assess the data management capabilities required to support AI agents and explore the importance of investing in an intelligent data layer with data agents that automate and accelerate key data processes.