娇色导航

Currently, AI agents are largely conversational. You prompt it, and it spits back an answer. While context windows have grown and integrations with external documents are improving, AI is still limited in how it interacts with the real world. However, this is poised to change dramatically as AI agents become more aware of application programming interfaces (APIs).

“APIs are going to be central to getting real value from agentic AI,” says Rebecca Fox, group 娇色导航at cybersecurity consultancy, NCC Group. “APIs are the glue — without them, agentic AI can’t effectively combine or orchestrate processes across different systems.”

Agentic AI promises to automate real business workflows, but only if it can connect to tools like CRMs, calendars, Slack, and payment systems. Since APIs are already the lingua franca of digital infrastructure, they’re the natural bridge to make agentic AI truly actionable.

“We see APIs as the cornerstone of agentic AI,” says Doug Gilbert, 娇色导航and CDO of Sutherland Global, a digital transformation services company. Connecting to external systems via APIs is critical to enabling enterprise automation. “Through structured function-calling mechanisms, agents autonomously invoke APIs to access data, execute tasks, and drive workflows,” he adds.

Stringing together multiple API requests across systems brings agentic workflows within reach. While CIOs underscore the importance of APIs, they also acknowledge the path forward still faces gaps — especially in documentation, governance, and complex access control — that must be addressed to achieve true AI-to-API interoperability.

APIs: integral to realizing agentic AI

APIs play two key roles in enabling agentic AI. “First, agents consume APIs to access tools and data in order to autonomously perform tasks,” says Mark O’Neill, distinguished VP analyst at Gartner. “Second, agents themselves may have APIs in order to instrument their actions.”

API access also goes beyond . It allows agents and their underlying language models not just to retrieve information, but perform database mutations and trigger external actions.

This shift allows agents to carry out complex, multi-step workflows that once required multiple human touchpoints. “AI-ready APIs paired with multi-agentic capabilities can unlock a broad range of use cases, which have enterprise workflows at their heart,” says Milind Naphade, SVP of technology and head of AI foundations at Capital One.

In addition, APIs are an important bridge out of previously isolated AI systems. “At the moment, services and platforms tend to push their own embedded AI capabilities,” says NCC Group’s Fox. “But realistically, businesses will need to connect AI across multiple applications and data sources through secure APIs.”

Compared to brittle workarounds like screen scraping or connector-based integration, APIs offer a far more robust foundation for automation. “I see APIs as critical to enabling agentic AI within the enterprise, allowing seamless real-time connectivity between diverse data sources, which is required for these agentic solutions to fulfill their value to our business,” says Joel Chaplin, 娇色导航at data management company Precisely.

How APIs make AI agents actionable

Early examples show agents autonomously managing calendars, retrieving emails, and summarizing meetings via APIs. But that’s just the beginning. From healthcare to insurance, and logistics to customer service, transformative agentic AI use cases are starting to emerge.

“By combining LLMs with robust tool integration, APIs enable agents to act as operational hubs,” says Sutherland’s Gilbert. In insurance, for instance, APIs can inform autonomous claims processing engines that extract data from external documents, validate claims against policy terms, detect fraud, and process outcomes with minimal human input.

AI agents can make unprecedented optimizations on the fly using APIs. Gartner reports that PC manufacturer Lenovo uses a suite of autonomous agents to optimize marketing and boost conversions. With the oversight of a planning agent, these agents call APIs to access purchase history, product data, and customer profiles, and trigger downstream applications in the server configuration process.

“The real transformation will come in areas like finance, warehouse management, logistics, and scheduling, where workflows are complex and traditionally hard to optimize,” says Fox. APIs could even reduce the need for bloated ERP platforms by replacing them with specialized services, cutting costs and complexity.

Plus, access to external APIs can enrich AI applications with relevant data and functionality beyond what an LLM can do alone, says Brian Glass, 娇色导航at Transcend, a health company focused on peptide therapies. Transcend is experimenting with Salesforce’s AgentForce to improve customer interactions, and their agents use APIs to validate orders for controlled substances, confirm eligibility, and flag potential side effects. Transcend is also exploring how API-driven agents can support service reps with real-time access to client, sales, demographic, and lab data, all while staying HIPAA-compliant.

New user-facing experiences are also emerging. Take for example Capital One’s Chat Concierge, an AI chat agent designed to improve the car-buying experience. Based on the open-source Llama model, it leverages additional APIs to perform actions, like comparing vehicles and scheduling appointments with dealerships.

“Many enterprise workflows can be reimagined as a specific instance of this multi-agentic conversational AI workflow technology,” says Capital One’s Naphade. He foresees similar patterns emerging across customer service, travel planning, analysis, and more.

Another area is project management. At Precisely, they’re working on capturing meeting notes and autonomously updating project trackers and dashboards via internal APIs. Precisely’s Chaplin recognizes the high potential of using API-enabled AI agents to improve employee experiences, as well as detect and remediate cybersecurity anomalies.

“We’ll see a wave of automation as agents begin taking on tasks across finance, logistics, and customer experience,” says Rory Blundell, CEO of Gravitee, an event-native API management platform. In DevOps specifically, agentic AI will likely help teams manage infrastructure by interacting with both synchronous APIs and event streams to make decisions, collaborate, and act autonomously.

Positive business outcomes using APIs

The use of APIs in agentic AI can be directly correlated to financial gains. For instance, the Lenovo architecture, which uses a swarm of AI agents to call APIs, has led to a higher conversion-to-order rate for AI-generated configurations compared to human-generated ones. O’Neill adds there’s also a business opportunity for monetized APIs that are appropriately tailored for agentic experiences.

But the bigger wins will likely be increased operational efficiency and cost reduction. As Fox describes, this stems from a newfound best-of-breed business agility. “When agentic AI can dynamically reconfigure business processes, using just what’s needed from the best-value providers, you’ll see streamlined operations, reduced complexity, and better overall resource allocation,” she says.

The outcomes for consumers are significant. “It’s more than just ‘let’s have a conversation,'” says Transcend’s Glass, describing how API access allows agents to guide users toward outcomes rather than simply answering questions. With APIs, AI agents can surface personalized data and talking points based on previous interactions, leading to safer and more productive experiences.

In medicine, precision, accuracy, and response time are critical, making reduced feedback loops highly important. “These systems can uncover things better than humans can, accelerating manual checking,” says Glass. “If an API can help us be more precise, then we’re all for it.”

Others agree that agentic AI can slim timelines and support risk assessments thanks to API access. According to Gilbert, insurance firms using agentic AI have cut claims processing times by 60% and costs by 30%. And APIs are set to further streamline workflows around onboarding, claims processing, and customer support.

Interoperability barriers must be solved

Still, there are big blockers to API interoperability with AI agents, one of which is legacy procurement processes. “Traditional APIs bring significant friction for AI agents because they were designed for human developers to access,” says O’Neill. API integration workflows that require speaking with a sales team or hide documentation behind a sign-in barrier are not well-positioned.

But walled gardens might exist for legitimate reasons. “Vendors often aren’t incentivized to fully open APIs because it potentially undermines their business model,” says Fox. Beyond that, ensuring seamless interoperability requires standardized, robust APIs that can handle frequent changes, she adds.

“The seamless integration of AI agents into enterprise ecosystems faces several key challenges,” says Gilbert. He cites problems with LLMs themselves, a lack of clear governance frameworks, and uncertainty with ongoing AI regulations. But a major blocker is integration with legacy systems. Outdated infrastructure with fragmented or poorly documented APIs can significantly slow down the integration process, he says.

Standards for API documentation are also nascent in many organizations. According to Enterprise Management Associates (EMA), only around 10% of organizations fully document their APIs. Even more troubling, a 2024 study from API monitoring company APIContext found 75% of production APIs diverge from their OpenAPI definitions. That means most APIs are either inaccurately described or lack proper definitions.

Interoperability hurdles will also hinge on a strong core data strategy — something not fully realized in many enterprises. “One of the biggest challenges facing enterprises is agentic AI’s need for vast and varied datasets, such as unstructured data like emails, documents, and videos, which are more difficult to validate and govern,” says Chaplin. “The state of data organization is varied in most enterprises. Rapid progress is being made, but true readiness for API-driven automation at scale remains aspirational for many.”

Precisely has taken this to heart, preparing its data for AI agents by centralizing data, applying metadata tagging, classification, and embedding techniques to make the content machine-readable and semantically rich. “It’s about turning raw content into structured knowledge that AI can reason over safely and effectively,” says Chaplin.

Security risks of agent-to-API connections

Opening up more API access to autonomous systems poses novel API security risks. This stems in part from the fact that agents are built on non-deterministic language models, which behave unpredictably. When acting autonomously or on behalf of other agents, their intent can become muddled.

“Security is probably the biggest headache,” says Fox, underlining the authentication and authorization implications. Given AI’s propensity to hallucinate, trusting it to respect granular permissions, such as who should have access to what, why, and when, presents a challenge. “Missteps here aren’t just inconvenient, they can expose sensitive data or create compliance nightmares.”

Nearly 95% of API attacks originate from authenticated users, according to 2025 data from Salt Security. Many hacks arise when an attacker simply modifies a user ID in an API request to obtain unauthorized information. This means that encouraging more API usage by AI agents could inadvertently amplify pre-existing access control gaps.

“Much of the current thinking about AI security focuses on OAuth,” adds O’Neill. But, OAuth 2.0, the ubiquitous API authorization framework, can only go so far. An agent may have the correct API authorization directives, but drifts its privileges at runtime, he says.

Without the proper safeguards, agentic systems risk accessing or misusing sensitive data, communicating with untrusted platforms, or experiencing prompt injection and input spoofing attacks, says Gilbert. “Unrestricted access could result in the unauthorized sharing of confidential information and an expanded attack surface.”

To respond, Sutherland is taking a multi-step approach. This includes using role-based API gateways, fine-grained access control, strict input validation, and strong observability practices to aid auditing and enemy detection. Overall, traditional endpoint protections won’t cut it — enterprises will require intent-aware, scalable governance frameworks to support AI autonomy, says Gilbert.

Standards required for API-to-AI integration

Connecting AI agents with APIs at scale will hinge on well-adopted industry standards. The primary one is MCP, the open-source protocol proposed by Anthropic. Since its release last year, MCP has spawned a large ecosystem of servers to connect AI agents with external data, tools, and APIs.

But while MCP adoption has skyrocketed in recent months, it’s not the end of the story. “We’re also excited about Google’s Agent-to-Agent protocol (A2A) to facilitate communication within an agentic AI system,” says Blundell. He expects solutions and frameworks that support these protocols to solve a lot of the agent-to-API connection challenges.

Others highlight the role of vendor-neutral standard specifications. O’Neill points to the Arazzo specification, a standard way to define sequences of API calls as well suited AI agent use cases. Using Arazzo, API providers can suggest pre-determined multi-step flows to agents to avoid randomness.

“Open standards such as OpenAPI, OAuth2, and GraphQL are also essential, ensuring secure and scalable integration across enterprise environments,” adds Gilbert. Thankfully, with new AI frameworks emerging, the onus won’t be on enterprises to string this all together themselves using open standards.

From the security standpoint, OAuth is an essential starting point, says Fox. She adds that more advanced security frameworks, particularly around zero-trust architectures and granular identity management, will become indispensable.

APIs turn observers into doers

APIs are widely seen as a linchpin for evolving agentic AI, but the current landscape is far from perfect. Enterprises still face challenges like inconsistent data practices, fragmented standards, , and rising security concerns. As a result, some leaders approach the space with cautious optimism rather than full adoption.

“Because the technology is moving at lightning speed, the focus should be on creating a connected, secure, and scalable environment where AI can thrive responsibly,” says Chaplin. For him, unlocking real value from agentic AI will require stronger governance at the data layer, including clearer policies for categorizing, defining, securing, and monitoring enterprise data.

Still, these roadblocks shouldn’t stall momentum. APIs will be essential to move autonomous agents beyond basic conversation and into meaningful, real-world execution. And, over time, organizations that master AI-to-API orchestration will likely outperform peers.

“APIs are the lifeblood of agentic AI,” says Blundell.” Without API access, agents remain observers rather than doers.”