Traditional database management has reached a tipping point. As data volumes skyrocket and systems grow increasingly complex, even the most seasoned administrators struggle to meet the operational demands of modern environments. Relying solely on manual oversight or reactive AI assistants is no longer enough.

From Breaking Point to Breakthrough - Enter Agentic AI, autonomous, intelligent agents designed not just to assist with database operations, but to actively manage, optimize, and adapt them in real time. These systems monitor conditions continuously, anticipate potential issues, choose the best tools for the task, and adjust strategies on the fly, keeping databases efficient, secure, and resilient.

Understanding Agentic AI
 There’s no single universally accepted definition of agentic AI. Some use it interchangeably with AI agents, while others make a clear distinction between the two. But regardless of the terminology, most agree on its core traits.

At its heart, agentic AI describes systems that can operate with a level of autonomy, able to make decisions and take actions in pursuit of specific goals. Unlike traditional AI, which typically waits for explicit prompts before producing results, agentic AI can assess a situation, plan a strategy, and execute multiple tasks, often in parallel. These systems choose how to accomplish their objectives, selecting the right tools and managing their own internal processes along the way.

Agentic AI System Architecture

Agentic AI isn’t just one piece of tech, it’s a design approach that gives AI systems more independence than traditional models. While it looks different depending on the application, most agentic setups use multiple large language models (LLMs) that talk to each other through prompts, tap into external tools, and handle reading and writing data sources. They often work in parallel rather than one step at a time, making them feel more like a network of collaborating agents than a single, isolated system.

Multiple roles assignment

In most cases, an agentic AI system isn’t just one agent working in isolation, it is a coordinated team of agents either from same model or different ones working in collaboration. Often, each agent is given a specific role.

For instance, one agent might play the role of a task manager, breaking a big problem into smaller pieces and assigning each piece to other agents. Those agents tackle their individual tasks and send the results back to the manager, who reviews and combines them into a final answer.

This modular setup makes it possible to take on more complex workflows than a single model could handle alone. Because many tasks can run at the same time instead of one after another, the system feels faster and more responsive.

Tool Calling

For agentic AI to go beyond simply generating text, it needs to use external tools. These might be APIs for pulling in live data, databases for storing and retrieving information, or file systems for reading and writing documents.

Frameworks like LangGraph, LangChain and LlamaIndex make this connection much easier, letting AI models tap into databases, pull search results, work directly with other software, hand-off data/results between agents to optimize the end results.

Concurrent Actions

Agentic AI often works concurrently, with several agents handling different parts of a problem at the same time, giving it a decentralized, network like feel.

Picture an AI troubleshooter - one agent pulls relevant documents/ground-truth from a vector store, another examines logs, a third dives into a specialized tool to pinpoint the root cause, and a fourth drafts the problem summary. By dividing and conquering, these independent agents can troubleshoot issues faster and more efficiently than a single model working alone.

Agentic AI in the IBM Db2 Intelligence Center

Do you know what the life of an enterprise DBA really looks like? In a typical large enterprise, there are over 20 DBAs, each spending nearly 50% of their time just monitoring systems, tuning queries, maintaining performance, and managing recovery tasks. That's thousands of hours spent on repetitive operations that could be optimized.

Now imagine reclaiming 90% of that time, because that's exactly what our agentic solution delivers. Our AI agents, rather than relying on hardcoded functions and path, it autonomously determines the optimal tools and parameters in real time, adapting its behavior to the unique demands of each problem it solves. It is designed to act as an intelligent co-pilot, it learns the environment, pinpoints bottlenecks, and provides actionable insights.

The results? A paradigm shift.

DBAs are no longer confined to firefighting and routine maintenance. Instead, they evolve into strategic enablers of innovation, supported by intelligent co-pilots that autonomously handle the operational load. These AI agents continuously learn, adapt, and optimize, freeing up time, reducing risk, and unlocking new possibilities for data-driven transformation. With this shift, enterprises gain more than just efficiency, they gain agility, resilience, and insight. DBAs become architects of performance, stewards of data intelligence, and catalysts for business growth.

Architectural diagram

Db2 Intelligence Center is a web-based interface that helps you understand, govern, and optimize your data within IBM Db2. It provides intelligent recommendations, insights into data usage, and tools for automating data quality and performance tuning.

Agentic AI Assistant is built on top of watsonx orchestrate platform, which provides a framework of guardrails to ensure reliability,  security and responsible usage. Db2 Intelligence Center (IC) sends agent-mode requests to the FastAPI-based applications, which in turn invoke the  LangGraph-powered agents. These agents leverage Redis as an in-memory state manager to maintain context awareness, and utilize MCP-based tools to enhance their capabilities. Depending on the use case, the tools retrieve metrics from multiple sources: the Vector Store containing ingested Db2 Documentation, the IC repository database for historical data, and the live database instance for real-time metrics in an agentic manner. This multi-agentic system leverages wx.ai models for inference, enabling intelligent decision-making and contextual responses.

Overall Db2 Intelligence Center Agentic Workflow

When a user submits a request in natural language, it first lands in the hands of the Agentic supervisor, the central coordinator of the workflow. If the request is unclear or ambiguous, the supervisor routes it to an Enhancer Agent. This agent’s job is to rephrase the question so it’s sharper, more precise, and ready for accurate processing before passing it back to the supervisor.

From there, the supervisor decides whether the query is a generic search-and-analysis request or specific to a database instance.

If it’s instance-specific, like Why is my query running slow? or Why is my job hanging?, the Performance Supervisor agent takes over. This agent passes the baton to the Performance Identifier, which conducts a deep-dive investigation. It leverages specialized analyzers such as:

·      Queuing Analyzer - Detects resource related job scheduling bottlenecks.

·      Spilling Analyzer - Checks for disk spills caused by memory constraints.

·      Statistics Analyzer – Analyzes runtime statistics and inaccurate cardinality estimates.

·      Query Plan Analyzer - Examines the execution plan to pinpoint inefficiencies.

Each of these tools contributes a piece of the puzzle, helping to zero in on the true cause of the slowdown.

Finally, once the analysis is complete, all findings are handed to the Final Answer agent, which compiles a clear, detailed, and well-formatted response for the user, transforming complex technical diagnostics into actionable insights.

Db2 Agentic AI - Your Key to 10x Productivity

When it comes to managing and optimizing Db2, there’s a lot going on behind the scenes. From diagnosing performance bottlenecks to helping you explore your data, Db2’s Agentic AI capabilities work like a team of specialized experts, each focused on a specific challenge.

Roadmap (Sept 30^th GA)^*1

To kick off our Db2 modernization powered by Agentic AI, we are productizing the following use cases to help customers get started on this journey.

Troubleshooting Agents

Think of these as your first responders, always on the lookout for issues before they impact your workload.

·      WLM Queuing Agent - Spots workload management queue buildups, explains why they’re happening, and recommends actionable fixes.

·      Disk Spilling Agent - Detects when queries overflow memory to disk, pinpoints the root cause, and helps you prevent it from recurring.

·      Change Monitor Agent - Tracks configuration or workload changes and correlates them with any performance drops.

·      Health Check Agent - Runs proactive diagnostics across workloads, flagging potential issues before they escalate.

Search & Analysis Agents

These are your researchers, analysts, and interpreters, making Db2’s knowledge and your data more accessible than ever.

·      Agentic RAG Agent - Context aware retrieval-augmented generation with Db2 expertise to pull answers from documentation, runbooks, and support pages.

·      Deep Research Agent - Performs multi-step reasoning to untangle complex Db2 topics and deliver clear, concise insights.

·      Text-to-SQL Agent - Lets you query and debug your data using plain English, no SQL expertise required.

·      Conversational Telemetry Agent - Allows you to have a natural conversation about monitoring metrics and your data model.

Next (post Sept 30)^*1

The next phase focuses on broadening agentic use cases and extending the reach of Agentic AI.

·      Next set of Db2 use cases

o   Advanced query tuning + impact analysis

o   Lock/Deadlock detection analysis

o   Resolve locking problems including lock waits, timeouts and deadlocks.

o   Resolve compression + table skew issues

o   Resolve blocking transactions (long lived transactions holding up logs)

o   Time spent analysis

o   Anomaly detection (of key KPIs)

o   Regressed SQL detection (variant of anomaly detection)

·      Integrating with the Db2 Bridge, a tool used to move data between Db2 databases. By combining this data mobility with Agentic AI, it will be able to analyze, optimize, and even automate cross-database operations.

·      SRE use case for faster detection and resolution of issues, reduced manual toil for SRE teams, improved system reliability and uptime, predictive prevention of performance degradations.

Future (Vision) ^*1
 The long-term vision is to make Db2 a truly self-managing and self-optimizing database -evolving from a system you manage, to a system that actively partners with you. Imagine a future where:

·      Self-Tuning Queries - Agents automatically rewrite poorly performing queries, optimize access paths, and recommend schema adjustments without waiting for manual DBA intervention.

·      Workload-Aware Scaling - Db2 anticipates workload spikes (e.g., end-of-quarter reporting, seasonal demand) and automatically adjusts resources across on-prem and cloud deployments to maintain consistent performance.

·      Cross-Db2 Migration Intelligence - With Db2 Bridge as the backbone, agents orchestrate seamless movement of data across Db2 environments, automatically validating integrity, performance, and compatibility along the way.

·      Autonomous Incident Management - Agentic AI integrates into SRE workflows, detecting anomalies before they escalate, simulating potential fixes, and safely executing resolutions, all while generating a clear explanation trail for teams.

·      Continuous Learning & Prediction - Agents learn from historical workloads, predict future bottlenecks, and preemptively recommend corrective actions to avoid downtime or regressions.

*1: IBM’s statements regarding its plans, directions, and intent are subject to change or withdrawal without notice and at IBM’s sole discretion.

About the Authors

Ashok Kumar - IBM Program Director, Hybrid Data Management
Ashok is a Program Director in Hybrid Data Management with 15+ years of experience leading Db2, BigSQL, and Data Virtualization across on-premises and cloud platforms. He is driving the modernization of Db2 with innovations such as the Db2 AI Optimizer, Vector Storage, and Agentic AI. Ashok has also contributed to open-source distributed engines like Presto/Prestissimo, which power IBM’s Wxd and Lakehouse solutions.

Krishna Guntuka is a Software Engineer with 8+ years of experience in development on various Db2 products, currently leading Agentic AI story for Db2 Database Assistant, his work focuses on different parts of these products, helping to make them work better and more efficiently.

David Kalmuk - IBM Distinguished Engineer, Analytical Databases

David Kalmuk is a Distinguished Engineer & Master Inventor and Chief Architect for Analytical Databases at IBM. David has contributed to the development of numerous technologies in Db2 over the years including BLU Acceleration, Workload Management, Monitoring, as well as much of Db2’s Processing and Communications architecture. He is currently leading efforts focused on evolving Db2's Cloud Native Warehousing and Lakehouse Architecture.

Satya Krishnaswamy - Director, Hybrid Data Management Development
Satya is responsible for driving the HDM strategy, development, and maintenance of software deliverables which contribute to the success of their software brand and/or overall software portfolio. He develops product strategy and product roll-out, makes investment decisions on new and existing software product features and functionality, support clients and client engagement teams and is also responsible for delivering defined business results such as growing revenue, increasing market share, reducing operational costs, and improving customer satisfaction as well as improving the climate of their organizations and building organizational capacity.