6 agentic knowledge base patterns emerging in the wild
AI agents have become the software industry’s latest fascination. Backed by large language models (LLMs), this new class of AI is unlocking data-driven decision-making and autonomous actions, transforming enterprise software practices and business workflows in the process.
However, it wasn’t always this way. According to Ajay Prakash, a senior staff software engineer at LinkedIn, AI agents initially faced a major gap. “Out of the box, AI coding agents weren’t effective,” Prakash tells The New Stack. They lacked context and awareness of internal systems, frameworks, and practices, he adds.
“Out of the box, AI coding agents weren’t effective. They lacked context and awareness of internal systems, frameworks, and practices…”
Agentic knowledge bases have emerged to close that gap. These systems allow AI agents to surface institutional data, runbooks, tools, project histories, and other context, helping them operate more effectively to deliver more consistent, verifiable outcomes.
As Anusha Kovi, a business intelligence engineer at Amazon, tells The New Stack, “Organizations are definitely building these, though it rarely looks like one centralized knowledge base product.” Instead, they’re often materializing as purpose-built layers to enforce accountability within specific domains.
Agentic knowledge bases are still taking shape, and their architectures, data sources, and use cases vary widely. Some are purely internal, while others are embedded within products or exposed to partners. What they have in common: coalescing organization-wide working standards. Below are six emerging patterns.
1. The playbook for coding assistants
The foremost use case for agentic knowledge bases in software development is context engineering. A prime example is LinkedIn’s knowledge base for AI agents, which acts as a source of truth for coding style and conventions, bringing platform consistency to AI-driven development. But it goes beyond enforcing style — it governs how agents act.
“Our knowledge base at LinkedIn enables AI coding agents to assist with company-specific tasks by providing tools to access internal systems and playbooks,” says LinkedIn’s Prakash. These playbooks encode rules, conventions, procedures, and verification steps.
One playbook focuses on debugging. “The agent proactively gathers relevant context,” says Prakash. “It fetches ticket details, pulls relevant logs, searches historical tickets, identifies related code paths, and classifies the likely team owners, automatically.”
More than gathering information, the playbook directs comprehensive operations. “The agent can apply the fix, run validation, and create a pull request with the original ticket linked, closing the loop from bug report to resolution,” says Prakash.
LinkedIn calls its framework contextual agent playbooks and tools (CAPT). It surfaces organization-wide instructions and dynamically exposes tool integrations and playbooks using Model Context Protocol (MCP). Interestingly, our playbook architecture anticipated what the industry later standardized as agent skills,” adds Prakash.
2. The integration knowledge center
Another use case is standardizing enterprise integration knowledge. Integrations are notoriously challenging to maintain, as fields change or contracts drift over time, making them brittle.
Adeptia has built an AI knowledge base as part of its data automation platform, empowering AI agents with institutional integration patterns and on-the-ground context. As Tim Bond, chief product officer at Adeptia, tells The New Stack, the knowledge base helps agents gradually understand how systems are used in practice.
“Agents interact through retrieval and augmentation,” says Bond. They first start with institutional knowledge, such as schemas, known integration patterns, and compliance requirements. Then, to tailor responses, the agents query situational context, such as prior conversations, workflow configurations, custom system mappings, and domain-specific terminology.
An example could be informing an AI agent how to automate a Salesforce-to-NetSuite integration, adds Bond.
With access to an AI knowledge center, agents can produce more valid, less generic integrations. “Agents become increasingly helpful over time, require less repetitive clarification, and handle more complex tasks,” says Bond.
3. The multi-agent home base
Agentic knowledge bases are also being designed to standardize multi-agent operations at scale. Organizations like R Systems, an IT service management company, are exploring knowledge bases as a foundation for such intelligent automation.
Neeraj Abhyankar, VP of data and AI at R Systems, tells The New Stack that an agent knowledge base serves as the company’s brain. “It gives every agent the same rules, voice, and playbook so they don’t improvise policy on the fly,” he says.
“We’ve implemented solutions that combine vectorized document repositories, semantic search, and retrieval-augmented generation (RAG) to support multi-agent workflows,” adds Abhyankar.
At the heart of their knowledge base is practical data and know-how, including policies, escalation paths, redaction rules, schemas, and vocabulary, runbooks like step-by-step how-tos for common tasks, and tool manifests.
Key benefits include faster agent action, greater consistency at scale, and improved governance, since responses and actions are traceable. “When a rule or template in the knowledge base is tweaked, the improvement shows up everywhere at once,” says Abhyankar. “How we work is encoded in the knowledge base, not scattered.”
4. The shared well of business context
It goes beyond software disciplines — agentic knowledge bases also empower business workflows. For example, Epicor’s knowledge base provides institutional knowledge on business data and enterprise resource planning (ERP) to inform financial and customer support agents better.
As Arturo Buzzalino, group vice president and chief innovation officer, Epicor, a provider of enterprise resource planning software, tells The New Stack, “We’ve built a centralized, organization-wide knowledge infrastructure designed specifically to support AI-driven work.”
In addition to accelerating access to business information, the knowledge base guides new workflows. “We saw an opportunity to enable true agentic automation,” says Buzzalino. “That shift from ‘tell me this’ to ‘handle this’ is what pushed us to formalize the knowledge base.”
Internal data supporting the knowledge base includes support documentation, system guidance, ticket histories, ERP-related knowledge, historical implementation data, and financial data.
“One of the most impactful examples is our financial agent,” says Buzzalino. Epicor’s knowledge base streamlines answers to complex financial questions, such as ‘Show me X metric for last quarter’ or ‘What’s our performance in Y region?’
“The agent retrieves the answer directly from the financial knowledge base, without needing to create a ticket, look at a dashboard, or wait for someone else with the right expertise,” says Buzzalino.
Epicor is seeing similar gains within its professional services division. For instance, agents can reference implementation data from past projects to improve real-time support.
Overall, the results have been increased agility and higher-value work delivered at lower operational cost, says Buzzalino. “It turns latent organizational knowledge into accessible intelligence, making individuals dramatically more capable in their roles.”
5. The source of truth for data intelligence
Other agentic knowledge bases are similarly emerging to bring consistency to business intelligence and data engineering disciplines. In these areas, a knowledge base can help an AI agent resolve confusion when it encounters conflicting fields or redundancies in databases.
A big driver is metrics chaos. “Conversations like ‘Wait, my dashboard shows a different number’ basically disappear when the agent always pulls from the same governed definition,” says Amazon’s Kovi. “That alone is worth it.”
“What I’ve seen in practice is teams assembling purpose-built layers that agents query before taking action,” Kovi adds. Structurally, the backend is composed of machine-readable definitions of tasks, schemas, data quality rulesets, and incident runbooks.
“The knowledge base isn’t there to help the agent be creative. It’s there to keep it inside the lines.”
On the business intelligence side, Kovi notes that these often emerge as a semantic layer, powered by metric definitions, exact SQL logic, dimensional hierarchies, business rule exceptions, and report formatting standards.
Although it sounds tech-heavy, your typical agentic knowledge base is using pretty standard components. “Most of it is YAML configs, versioned markdown, and structured catalog tables exposed via API.”
According to Kovi, agent knowledge bases reduce the repetitive grind of incident triage in data engineering. Converging on a single, one true definition of a metric can avoid getting different answers from slightly different prompts for the same source.
“You’ve got three teams with three different SQL definitions of ‘revenue,’ and the agent just grabs whichever one scores highest in a vector search,” Kovi says. “The knowledge base becomes the thing the agent has to check before it writes any query.”
6. The MCP-powered capability layer
Arguably, new infrastructure around MCP, such as MCP gateways and MCP registries, takes agentic knowledge bases to their zenith, since they add a structured means for AI agents to access sanctioned capabilities, bringing governed power to what an LLM can automate on its own.
For instance, Vendia, an AI platform provider, has built an MCP gateway to construct agentic knowledge bases.
Tim Wagner, CEO and co-founder, Vendia, tells The New Stack that the MCP protocol is enabling a sea change in agent architecture. It unlocks LLM intelligence and simultaneously lowers the cost, delivery risk, and complexity of building agents that can access and eventually maintain AI knowledge bases.
Wagner adds: “Companies are starting to migrate from prompt stuffing techniques like RAG that require the company’s developers to manually search and compute results out of the knowledge base and instead let the LLM automate its own searching, retrieval, and results exploration.”
The revolutionary aspect is that, through MCP, he says, AI agent queries can be combined with all of a company’s other assets and services, including third-party software-as-a-service APIs, internal applications, and cloud and on-premises content management systems.
Alongside MCP servers, he’s seeing organizations store media and intellectual property assets, policy documents, manuals, and more within AI knowledge bases. The use cases vary across sectors and span both internal and external-facing workflows.
For Wagner, the benefits of an agentic knowledge base are primarily customer-based. “Usually, improved employee or customer outcomes are the driving force today, more so than automation or cost reduction.”
Agentic knowledge bases bring clear results
Many companies, platforms, and infrastructure providers are actively constructing AI knowledge bases for various purposes. The resounding effect is to govern access to data and capabilities and to reduce tribal knowledge in the process.
Bond shares that Adeptia’s approach of combining historical knowledge with situational context has reduced dependence on human support while maintaining accuracy.
After implementing an agentic knowledge base, LinkedIn has seen a 20% increase in AI coding adoption and measurable productivity gains in debugging and data analysis. “Issue triage time has dropped by approximately 70% in many areas,” says LinkedIn’s Prakash.
In data engineering, a knowledge base establishes an institutionally defined context to prevent an LLM from misinterpreting information or assuming database mechanics. “The knowledge base isn’t there to help the agent be creative,” says Amazon’s Kovi. “It’s there to keep it inside the lines.”
The groundwork for agentic AI
A 2026 Zapier study found that 25% of enterprises expect to achieve full-scale orchestration by 2026, in which AI functions as an operating system for the business. 43% anticipate reaching the agentic AI stage, defined by more autonomous linking of systems and workflows.
More AI agent development is underway, and knowledge bases will be a key component to support its maturity. Today’s early implementations demonstrate much potential as a foundation for future iteration.
That said, challenges remain: Keeping data fresh will require pipelines to continually update the knowledge base and ensure data quality. Experts also recommend using an open, standards-based approach, maintaining distributed ownership, establishing version control, and intentionally federating knowledge.
“The space is still evolving,” says R Systems’ Abhyankar. “Structured knowledge bases can empower AI agents to deliver reliable, context-aware assistance and lay the groundwork for broader enterprise adoption of agentic AI.”
