Kong’s 2025 year in review shows the rise of AI connectivity

At a glance

What this is: This is Kong’s year-end analysis of AI connectivity in 2025, with the key finding that agentic adoption is forcing organisations to treat APIs, LLMs, agents and event streams as a governed control plane rather than a simple integration stack.

Why it matters: It matters because IAM, API security and NHI governance now intersect at runtime, where machine-speed access decisions, token controls and policy enforcement need to keep pace with autonomous workloads.

By the numbers:

• 90% of enterprises are adopting AI agents, according to Kong research.
• 79% expect full production rollout within three years, according to Kong research.
• 72% of organisations are already implementing formal governance layers, according to Kong research.

Context

Kong’s 2025 year in review argues that AI connectivity is becoming a governance problem as much as an integration problem. The article’s core message is that enterprises are moving from human-paced API usage to agentic systems that generate traffic, requests and policy decisions at machine speed.

For IAM and security teams, the important shift is not the marketing language around connectivity but the control surface behind it. When APIs, LLMs, agents and event streams converge, the real question becomes how identity, authorisation and usage policy are enforced across short-lived, autonomous interactions rather than static application calls.

Key questions

Q: What breaks when AI connectivity is added without policy controls?

A: Teams end up with hidden trust edges, overbroad service access, and logs that do not explain which workflow triggered the action. That makes containment, review, and incident response much harder, because the organisation can no longer distinguish legitimate machine activity from accidental overreach.

Q: When should teams prioritise governance over broader agent rollouts?

A: Teams should prioritise governance before scaling agent use whenever the system can act on production data, invoke tools, or affect downstream workflows. If those boundaries are not defined first, every additional deployment makes the control gap harder to close and harder to explain to auditors or risk owners.

Q: How do teams know whether an AI gateway is actually improving control?

A: They should measure whether the gateway reduces unmanaged keys, improves audit completeness, and enforces consistent access rules across all providers. If teams still maintain ad hoc model credentials, cannot explain fallback behaviour, or lack prompt-level telemetry, the gateway is not yet functioning as a governance layer.

Q: What is the difference between API governance and AI gateway governance?

A: API governance focuses on standardizing how services expose and consume APIs, including versioning, authentication, rate limiting, and access control. AI gateway governance applies the same discipline to LLM calls, MCP tool access, and agent-to-agent communication. The difference is the traffic type and the controls it needs, especially token awareness, prompt security, model routing, and agent identity management.

Technical breakdown

Why AI gateways are becoming runtime control planes

An AI gateway is essentially a policy enforcement layer for AI traffic, sitting between consumers and the services, models or tools they invoke. In Kong’s framing, that means governing token budgets, applying security checks, supporting semantic caching and managing the operational cost of agentic workloads. For practitioners, the key technical point is that this is not just API mediation. It is a runtime control plane for machine-originated requests that may chain multiple downstream actions in one session.

Practical implication: Treat AI gateway policy as part of identity governance, not just traffic management.

MCP authentication and agent connectivity risks

The article’s MCP references point to a broader pattern: agent connectivity now depends on consistent authentication, predictable server behaviour and controlled workflow access. Where agents orchestrate tools and event streams, the attack surface expands to include identity delegation, session scope and trust between services that were not designed for autonomous callers. That makes authentication and authorisation failures more consequential, because a weak connector can become the route by which an agent reaches multiple systems in a single workflow.

Practical implication: Review connector authentication, delegated scopes and workflow trust boundaries before allowing agentic access.

Why AI fragmentation becomes an identity problem

Fragmentation appears when teams bolt together separate systems for APIs, models, agents and events without one governance model across them. The result is inconsistent policy, duplicated controls and unclear ownership of who or what can invoke which capability. In identity terms, this is a lifecycle and entitlement problem: every new connector introduces another place where access can be over-broadened, under-reviewed or left without revocation discipline.

Practical implication: Map every AI-connected component to an owner, an access boundary and a revocation path.

NHI Mgmt Group analysis

AI connectivity is becoming the governance layer that sits above modern machine-to-machine access. Once APIs, LLMs, agents and event streams are handled together, the enterprise is no longer governing isolated connections. It is governing a shared policy surface where identity, usage and risk are decided in motion. The practitioner takeaway is that AI connectivity now belongs in the same governance conversation as workload identity and privileged access.

Token budgets are an identity control, not just a cost control. When agentic systems consume model capacity and downstream services at scale, the budget decision influences which actors can proceed, how far they can chain actions and where usage is constrained. That makes budget enforcement part of authorisation design, especially when autonomous workloads can amplify small policy mistakes into large operational exposure. Practitioners should treat usage limits as a control boundary, not a finance setting.

AI fragmentation tax is really policy fragmentation debt. Kong’s description of fragmented AI infrastructure maps to a familiar governance failure: each new tool, connector or stream introduces another policy exception. Over time, that creates inconsistent identity assumptions across the stack, which is exactly how unmanaged access expands faster than review cycles can shrink it. The implication for practitioners is to reduce policy drift before it becomes structural.

Formal governance layers are becoming the minimum viable response to agentic scale. The article’s numbers show enterprises are already moving from experimentation to governance, which suggests the market is converging on runtime controls rather than point solutions. That direction aligns with broader NHI and agentic AI governance patterns: who or what is acting matters less than whether the access path is bounded, observable and revocable. Practitioners should assume governance is moving closer to execution time.

AI connectivity will force IAM teams to own cross-domain identity boundaries. The article ties together APIs, events, LLMs and agents because production systems now mix them in the same transaction path. That means IAM, API security and NHI governance can no longer be managed as separate programmes with separate owners. The practical conclusion is to design one access model that covers human, non-human and agentic callers where the workflow depends on all three.

What this signals

AI connectivity will push governance closer to execution time. As agentic systems begin to orchestrate APIs, models and event streams together, the old assumption that access can be reviewed after the fact becomes weaker. Practitioners should expect policy decisions to move toward runtime enforcement, where the control point sits at the moment of invocation rather than at the next certification cycle.

Runtime policy will matter more than integration completeness. In practice, the programme that can bound token use, connector scope and workflow escalation will be more governable than one that simply links more services together. The design target is not total connectivity, but controlled connectivity with clear ownership, revocation and observability across each non-human caller.

For practitioners

• Define an AI connectivity governance boundary Inventory the APIs, models, agents and event streams that participate in agentic workflows, then assign a single policy owner for each boundary and escalation path.
• Classify token budgets as access controls Set budget thresholds, approval rules and exception handling so that model usage limits are enforced alongside authorisation, not after the fact.
• Map connector authentication to delegated trust Document which connector uses direct service credentials, which uses delegated scopes and which can call downstream services without human intervention.
• Establish revocation paths for agentic integrations For each agent, tool and event integration, define how access is removed when workflows change, vendors change or a connector is retired.

Key takeaways

• AI connectivity is becoming a governance layer because agentic systems now drive API, model and event traffic at machine speed.
• The article’s own numbers show that adoption is already widespread, which means governance is catching up after experimentation has begun.
• Practitioners should treat token budgets, connector trust and revocation paths as identity controls inside the AI runtime.

Key terms

• AI Connectivity: AI connectivity is the enterprise control layer for governing how applications, users, and agents reach AI services. It combines routing, security, observability, and policy enforcement so AI interactions can be managed consistently across providers and use cases.
• AI Gateway: A control point that sits between AI applications and the models, tools, or data they call. In practice, it can authenticate requests, enforce policy, inspect runtime behaviour, and stop unsafe actions before they spread into connected systems.
• Token Budget: A token budget is the amount of text capacity allocated to a model response, including in some systems the hidden reasoning steps that occur before output. If the budget is too small, the answer may truncate, disappear, or fail without a clear error condition.
• Agentic workflow: An agentic workflow is a sequence of tasks executed by an AI agent with some level of tool access and decision authority. In security terms, the workflow matters because it can span multiple systems, identities, and permissions, which makes attribution and revocation harder than with ordinary automation.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.