How agentic AI is set to redefine enterprise APIs [Q&A]
The use of AI across modern enterprises in recent years has accelerated, with innovation at the forefront and APIs serving as the crucial enabler behind the scenes.
Now, agentic AI, capable of autonomous actions and decision-making, but this shift exposes several gaps in API documentation, drift in specifications and insufficient safety guardrails, all of which can lead to serious implications for organizations.
We spoke to Mayur Upadhyaya, CEO of APIContext, to discuss how agentic AI will impact enterprise APIs and the best practices organizations should follow to prepare their APIs for this shift.
BN: What is agentic AI, and how are enterprises beginning to adopt and apply it in practice?
MU: Autonomous agents operate independently to accomplish specific objectives. In practice, this enables them to perform complex tasks, streamline dynamic processes, enhance customer support, and strengthen data analysis for more informed decision-making all without direct human intervention. As a result, agentic AI adoption within enterprises has accelerated, with over 51 percent of organisations already using AI agents and another 35 percent expected to follow by 2027.
Agentic AI has risen in popularity by enabling organisations to automate and streamline routine processes such as customer support queries, real-time supply chain optimisation , and detecting anomalies in large datasets before they become critical issues. This allows teams to operate more efficiently, make faster decisions, and deliver more personalised and responsive services to customers and stakeholders.
BN: How will the rise of agentic AI change the way organizations design, document and manage their APIs?
MU: AI agents rely on APIs to interact with data sources, applications, and external services, enabling them to exchange information, trigger actions, and collaborate effectively across systems. Simply put, without APIs, agentic AI cannot function. With this in mind, organisations must carefully design, document, and manage their APIs in several key ways.
First, keeping API documentation up-to-date is critical. However, this remains a challenge in the current landscape as our research found that 89 percent of specifications had not been updated in the past six months.
Next, agentic AI needs good API documentation. Beyond missing documentation, most APIs have API drift, where the documented specification is different from the production API. Plainly, the documentation doesn’t reflect reality. If drift occurs, AI may fail to function as intended. With this in mind, it is vital to keep documentation fully up to date.
As organisations begin to onboard autonomous agents, managing API usage will become more complex. The Model Context Protocol (MCP), an emerging open standard, provides a consistent interface that allows AI agents to access and orchestrate APIs efficiently. While MCP simplifies integration at scale, it also drives exponential API consumption, creating new operational and performance challenges for downstream systems.
BN: What are the risks or consequences for organizations that fail to prepare their APIs for agentic AI?
MU: If enterprise APIs are not prepared for agentic AI, organizations face several risks. AI agents rely on accurate, up-to-date API specifications to function reliably. Without proper documentation, agents may call deprecated endpoints or misinterpret functionality, breaking workflows and undermining automation efforts.
Organizations also need to consider how API drift will impact autonomous agents. Given that 75 percent of APIs show at least one nonconformant endpoint, API drift is already widespread. For AI agents, unpredictable or outdated APIs can cause failed calls, inefficiencies, or incorrect outcomes at scale which are issues that become unmanageable.
Finally, Model Context Protocol (MCP) standardizes how AI agents access APIs, enabling orchestration at scale. However, this efficiency can also dramatically increase API consumption, straining downstream systems. Organizations unprepared for this surge risk service degradation, instability, and higher operational costs.
In short, failing to update and align APIs with agentic AI exposes businesses to inefficiencies, outages, and scalability challenges that can quickly undermine competitive advantage.
BN: What best practices should organizations adopt now to ensure their APIs are ready for autonomous agents?
MU: To ensure APIs are prepared for autonomous agents, organizations should start by treating OpenAPI Definitions as living contracts, kept accurate and continuously updated within development workflows. This reduces the risk of API drift and ensures agents can consume APIs reliably. Specifications should include machine-readable hints, descriptive field names, and explicit constraints to help agents avoid invalid inputs. Formalizing business logic within specifications or companion documents further ensures compliance with organizational policies. Finally, deploying agent gateways or MCP servers provides a vital control layer, validating inputs, monitoring behaviour, and preventing misuse or overload as AI adoption accelerates.
BN: Where do you see the biggest opportunities and risks when it comes to agentic AI and enterprise API strategy over the next three to five years?
MU: Today, LLMs act as the ‘smartest interns ever’ -- incredibly knowledgeable and insightful, but often unable to interpret basic context about how and where to apply their knowledge. Agentic AI will raise the stakes for this dynamic, creating opportunities for well designed agents to excel at mundane tasks, and free up valuable human resources with much less busy work. But, in order to achieve those goals, we need guardrails that constrain these tools against avoidable errors, especially when they can extract and manipulate data, and make decisions without human oversight. Because LLMs and foundational AI models are evolving so quickly, monitoring the real infrastructure that they rely on is crucial to ensuring they first do no harm.
Image credit: [email protected]/depositphotos.com
