Get in touch
AI
API Management
  • 5 min read

From Pipes to Partners: API Maturity in the agentic era

From syntactic contracts to cognitive transparency. The seven stages of API maturity in the agentic era, and where MCP and A2A actually fit.

  • Published on
    28 mei 2026

Introduction

In the “AI Doesn’t Like Your API. How to Fix It” blog, we explored why AI does not automatically like your API.

The problem is not that APIs are broken.
The problem is that most APIs were designed for a different world.

They were designed for predictable application-to-application integration. They were designed around contracts, schemas, endpoints and predefined flows.

That remains valuable.

But AI-native and agentic environments introduce a new integration challenge.

Systems do not only need to exchange messages.
They need to interpret meaning.
They need to understand behavior.
They need to express intent.
They need to coordinate dynamically.
And they need to remain observable and trustworthy while doing so.

That is why API maturity needs to evolve.

Not because REST, gRPC, GraphQL or OpenAPI are suddenly irrelevant.

But because they only solve part of the problem.

They help systems connect.

They do not automatically help systems collaborate.

API maturity stages: from pipes to partners

One way to understand this evolution is to look at APIs through maturity stages.

Each stage adds a new layer of interoperability.

No formal contract. Manual integrations, hardcoded behavior, CSVs over FTP, email APIs, etc.

→ Raw Pipelines & Ad-hoc Connectivity

Formal interface contracts like WSDL, UDDI, OpenAPI. Focus is on structure, transport, and protocol correctness.

→ Standardized Schemas & Grammars (JSON, XML)

APIs behave in expected ways over time: retries, long-running Ops, lifecycles, eventual consistency, etc. Often under documented. REST(ful) partially covers this.

→ Predictable State Changes, business behavior & Side Effects

Shared understanding of meaning: domain models, concepts, context. Needed for AI to reason safely.

→ Shared Concepts, Ontologies & Knowledge Graphs

Goal-driven interfaces: “book delivery,” “assign task.” Abstracts low-level operations into business intent. MCP enables intentional operations, but doesn’t solve interoperability. OpenAPI – Arazzo proposes a specification that could generate MCPs.

→ Pragmatic Context & Goal Execution

Adaptive, feedback-aware APIs. Co-evolving contracts that support negotiation, learning, and dynamic orchestration with AI agents. A2A (Agent-2-Agent) is positioned as the integration standard to enable this maturity level.

→ Multi-Agent Workflow Orchestration & Process Alignment

Shared perception, fluid adaptability to unprecedented edge cases, and Explainability / Trust. It bridges the gap between machine automation and human cognitive processing, ensuring that as systems act autonomously, they remain cognitively transparent to the humans managing them.

→ Shared Mental Models, Explainability, Human-Machine Teaming

A2A protocol as an evolution

A2A (Application to Application) interaction, is evolving. In AI native and adaptive environments, it needs to move beyond syntactic and behavioral maturity toward a truly collaborative stage.

Today, however, most A2A implementations still sit somewhere in between.

Where A2A typically sits today

Syntactic: Interfaces are clearly defined through XML, JSON, and API contracts. Systems can exchange data reliably.

Behavioral: Flows are orchestrated through BPMN or event-driven patterns. Systems coordinate actions, retries, and processes.

But something is still missing.

Systems can talk. They can coordinate. But they do not truly understand.

What is missing is semantic clarity and intent awareness. Closing that gap is what moves A2A from coordinated interaction to true collaboration.

MCP and Agent-to-Agent: useful, but not magic

Two concepts are becoming increasingly important in this evolution: MCP and Agent-to-Agent interaction.

MCP helps an agent use tools and context. A2A helps agents collaborate with other agents.

MCP is therefore most relevant to intentional API consumption and tool access. A2A is most relevant to process and collaborative interoperability.

Both still depend on semantic quality underneath. So MCP and Agent-to-Agent protocols are not replacements for good API design. They are accelerators for landscapes that already invest in semantic clarity, behavioral predictability and architectural governance.

What that evolution looks like

The evolution toward collaborative APIs and agentic interoperability has several dimensions.

Instead of rigid, predefined flows, applications and agents express intent. They define outcomes and work toward them, adapting along the way.

Shared vocabularies, units, and constraints become essential. Systems no longer just parse data. They interpret meaning.

Like human teams, applications and agents may need to adjust their behavior depending on roles, context, and capabilities. Think: “I can handle this if it’s urgent,” or “I need that dependency before continuing.”

Agents act based on a common understanding of current state, goals, and constraints rather than isolated API calls.

The intent was always collaborative

Unlike traditional APIs, agent protocols were designed from the start with collaboration, semantics, and intent in mind.

Agent2Agent protocols are not merely another API format. They start higher in the interoperability stack than traditional APIs because they are explicitly designed around task exchange, capability discovery, long-running coordination, and collaboration between autonomous agents. But they do not magically solve semantic, process, or cognitive interoperability. A2A gives agents a way to talk and coordinate. The quality of collaboration still depends on the meaning, constraints, policies, explanations, and shared context carried through that protocol.

But in practice, many implementations still ran into familiar limits:

  • they struggled with real world ontological alignment
  • they relied on brittle, hand coded interaction patterns
  • they lacked the adaptive reasoning and self explanatory behavior needed for open collaboration

So while A2A protocols may conceptually operate at the collaborative level, many real world implementations still fall short. Especially when the underlying enterprise landscape is not ready.

Summary

  • A2A (Agent-to-Agent) protocols were originally built with collaborative intent.
  • They’re the ideological predecessors to AI-native API thinking.
  • But their broader impact was constrained by limited adoption and by the difficulty of achieving semantic alignment in real environments.
  • With the rise of LLM based agents, those ideas now have a new chance to mature, supported by more powerful tools such as ontologies, embeddings, and context aware reasoning.

At the same time, MCP and Agent-to-Agent protocols do not magically solve interoperability. MCP can help agents use tools and context. Agent-to-Agent protocols can help agents coordinate with other agents. But both still depend on the semantic and behavioral quality of the integration landscape underneath.

That is why API maturity matters more in the AI era, not less.

The journey moves from raw connectivity to syntactic contracts, from predictable behavior to shared meaning, from goal-driven interfaces to multi-agent collaboration, and eventually toward explainability, trust and human-machine teaming.

 

“

“In other words, APIs are evolving from pipes to partners. And integration architecture is what makes that evolution governable, scalable and safe.”

Wim Paredis
Managing partner

Latest articles

Read all insights

Tap into the knowledge
of our community.

Back to the knowledge hub Get in touch