🇪🇸Spring I/O 2026

I just got back from Spring I/O 2026 in Barcelona, and, as usual, the event left me with a lot to process — in the best possible way. Spring I/O is one of those conferences that manages to feel both tightly focused and surprisingly broad at the same time. This year, the gravitational pull was clear: AI, agentic patterns, and the evolution of the Spring ecosystem (Spring 7.x and Spring Boot 4.x). From the hall track discussions, it is clear that organizations will be wrestling to answer a set of questions about how to build, test, and understand agentic applications running on the JVM.

The Workshop: Testing and Observing Agentic Applications with Spring AI

The week kicked off with a workshop I ran alongside Laurent Broudoux — from Postman / Microcks / Reshapr — focused on testing and observing agentic applications built with Spring AI. We had a great crowd in the room, and the conversation went deep fast. The workshop walked attendees through real patterns for asserting agent behavior, mocking LLMs' responses, testing and mocking MCP servers, and making the invisible visible through traces.

The workshop code that can be found here https://github.com/salaboy/spring-io-2026-workshop shows how to start simple with a Spring AI application with local/internal @Tools to then expand to use MCP to consume REST and gRPC APIs. The workshop ended with Dapr Workflows for building durable orchestrations and running all the components inside a Kubernetes cluster.

We had fun with Laurent preparing and running the workshop, so expect more content, talks, and workshops around Observability, OpenTelemetry, MCP servers, Reshapr, and of course, Microcks for mocking.

Spring AI 2.x and the Agentic Push

The buzz around Spring AI 2.0 was impossible to miss. The project is maturing fast, and the upcoming 2.0 release brings a serious set of new patterns for building agentic systems. Two things stood out to me in particular.

First, Spring AI 2.0 ships with first-class MCP (Model Context Protocol) client and server libraries built in. MCP is the protocol that has quickly become the standard way for agents to expose tools and for orchestrators to consume them. Having that integrated natively into the Spring ecosystem rather than bolted on as an afterthought is a big deal for teams already invested in the Spring Boot programming model.
Second, the project is actively addressing patterns across the emerging protocol landscape — MCP, A2A (Agent-to-Agent), and ACP (Agent Communication Protocol) are all on the radar. Running multiple protocols concurrently and understanding the observability challenges that come with that is genuinely non-trivial. When you have agents calling other agents through different protocols, tracing a single logical operation end-to-end becomes a real engineering problem.

Congrats to Christian Tzolov, Mark Pollack, and the rest of the Spring AI team for all the insights they shared over the conference.

If you want to learn about Agentic Patterns with Spring AI, you should check this repository: https://github.com/tzolov/voxxeddays2026-demo,

Observing AI: Semantic Conventions, OpenInference, and a Maturing Ecosystem

The OpenTelemetry Gen AI semantic conventions are starting to stabilize, and this is genuinely exciting. For the first time, we have a shared vocabulary for what an LLM span should look like — what attributes to capture, how to represent tool calls, how to model embeddings. Without agreed conventions, every vendor builds their own schema and you end up with a bunch of incompatible dashboards.

On that note, OpenInference from Arize AI deserves a mention. OpenInference is an open standard for capturing traces from AI applications — it sits on top of OpenTelemetry and adds the AI-specific semantics. Arize has been pushing hard on tooling here, and I think the approach of building on OTel rather than replacing it is the right call.

At its core, the Spring team's investment in Micrometer as the single abstraction over metrics, traces, and soon logs is paying off. Teams using Spring Boot get a coherent observability story without having to wire everything together manually. I want to thank Jonatan Ivanov and Tommy Ludwig for the very insightful session and the deep conversations on the observability roadmap.

The Arconia moment for observability

More observability conventions and challenges are popping up, and this is where I need to share my excitement about the Arconia project, introduced by my friend Thomas Vitale, who has built a bunch of features to improve the developer experience on top of the Spring Boot framework.

One of those improvements actually comes from expanding Spring's support to the Otel GenAI semantic convention and supporting the OpenInference conventions while at the same time providing dev services to bootstrap tools like Phoenix Arize using Testcontainers.

I will be writing more about this, as I believe that, with the accelerated pace in the agentic and observability space, you sometimes cannot wait for frameworks to catch up; we all need to collaborate to bring solutions to new challenges. I am really looking forward to contributing my findings back to the Arconia project.

Big congrats to Thomas for the announcement 🥳🥳🥳

Barcelona, the Community, and What Comes Next

Beyond the technical content, Spring I/O is just a genuinely very special conference. Sergi Almar put together something that feels like meeting friends who share the same passion as much as a conference. I ran into so many familiar faces and met a bunch of new people I hope to stay in touch with.

I am super hyped about Spring I/O 2027, as it will happen in Valencia for the first time ever!!

The energy around AI in the Spring ecosystem right now is real, and it's moving fast. Spring AI 2.0, the maturing observability stack, and the protocol convergence around MCP, Skills, A2A, and ACP are all happening simultaneously.

My personal takeaway matched what was shared at the conference: the Java ecosystem is strong because, with a flexible, agent-based programming model, organizations can build robust solutions that leverage all the enterprise integrations built over the last 20 years in the Spring ecosystem.

If you attended Spring I/O and want to compare notes, or if you're working on anything in the AI observability or Spring AI space, reach out — you can find me on Twitter/X @salaboy or LinkedIn.