Agentic data management starts taking shape as a tech option

Build or buy data management agents?

Enterprises looking to adopt agentic data management also have a choice to make between building their own agents or using ones offered by vendors. In a hybrid approach, they can also use vendor-provided development kits to create custom agents.

The build-vs.-buy decision "depends heavily on the maturity of the company's data engineering environment," said Sami Rahman, head of AI strategy and solutions at DataArt, a software development services and consulting company. For instance, he noted that companies built on digital products are often more inclined to develop their own agents and connect them to tools such as DBT for data transformation and the Apache Airflow workflow orchestration platform.

The DIY approach has a downside, though. Rahman said managing custom code can become difficult, especially when data governance requirements change. "This is one reason vendor capabilities are becoming more attractive," he observed. "They often include permissions and audit support, and some provide lineage capabilities as well."

Rahman thinks most organizations will likely deploy a combination of vendor-developed and homegrown agents -- something he's already seeing among early adopters.

"In practice, a mixed setup is common, with vendors handling routine platform work while custom agents are reserved for business rules or specialized workflows that are specific to the organization," he said.

What vendors are offering

Ronthal said vendors are providing prebuilt agents for specific data management and analytics tasks, such as data engineering or conversational analytics. Meanwhile, the agent development studios and kits offered by vendors often enable data teams to build their own agents "by simply describing what the agent should do," he said.

Examples of vendors that have introduced agentic data management capabilities include Acceldata, Alation, Ataccama, Google Cloud and Informatica.

Acceldata, for instance, provides purpose-built agents for functions including data quality management, data catalog development and data pipeline monitoring, failure detection and optimization. It also offers Agent Studio, which lets organizations create and deploy custom agents within the company's Agentic Data Management platform.

Similarly, Informatica offers a set of prebuilt Claire Agents for data management tasks such as creating data quality rules and building data pipelines, as well as an AI Agent Engineering cloud service with a no-code interface for building, orchestrating and managing agents within Informatica environments.

Ataccama sells One AI Agent, which can create and apply data quality rules, profile data and identify issues in datasets, while Alation CDE Manager provides a set of agents for managing and governing critical data elements used in financial reporting, risk management, regulatory compliance and other corporate processes. Acceldata, Alation, Ataccama and Informatica all launched their offerings in 2025.

Google Cloud offers out-of-the-box agents for data engineering, database provisioning and performance monitoring, and data science tasks such as exploring, cleaning and preparing data. They're part of a broader Agentic Data Cloud platform announced in April 2026 that also includes a Data Agent Kit for creating custom agents.

Agentic data management challenges

Agentic data management is beginning to move beyond proofs of concept and into production deployments, according to Rahman. But he added that adoption remains uneven and the technology faces challenges, such as the need to implement strong processes for managing agents.

"The underlying models are capable of handling useful parts of the workflow, but broader enterprise use depends on the operational structure surrounding them," Rahman said. "Organizations still need reliable evaluation frameworks, along with defined escalation procedures for cases that require human intervention."

He expects agentic capabilities to become standard functionality in modern data platforms over the next two years. As those capabilities become more widely available, the quality of the governance controls built around them will increasingly become the main point of differentiation for platform vendors and users alike, Rahman said.

"This will make agentic data management an important proving ground for enterprise AI governance," he noted.

Ronthal also pointed to governance as an important factor, describing it and a lack of cost management maturity through AI FinOps as the primary headwinds facing agentic data management in its early stages.

In addition, most agents are currently focused on a single task and haven't been built to collaborate with other agents, Ronthal said. "We are not yet seeing widespread deployment of complex multi-agent workflows."

He added that multi-agent optimization will rely on graph analytics, model routing and other advanced aspects, which will delay the full-scale maturity of agentic data management. Nonetheless, Ronthal expects the technology to mature relatively quickly.

Adoption strategies for data teams

Ronthal said initially focusing on well-defined deployments, with an emphasis on cost control, enables organizations to develop expertise with lower-risk agents before implementing ones likely to incur higher costs.

Rahman also cited an emphasis on early value as a wise approach in agentic data management deployments. "Most teams are focusing on areas where the scope is defined and the benefit becomes apparent fairly quickly, particularly as data volumes increase," he said.

Common starting-point use cases such as data quality remediation, schema mapping and pipeline monitoring provide a foundation layer for agentic data management, Rahman said. With that in place, organizations can then build agents to handle some of the work that happens before AI model training, such as profiling data sources and drafting data transformation rules.

But at this early stage of adoption, most organizations remain cautious about giving agents full control of data management workflows.

"The most common approach has an agent completing the first pass before an experienced data engineer reviews the output, makes any necessary adjustments and approves it," Rahman said.

John Moore is a freelance writer who has covered business and technology topics for 40 years. He focuses on enterprise IT strategy, AI adoption, data management and partner ecosystems.