Agentic Engineering & ToolingDeep Dive180min
Can AI Build a Jakarta EE Runtime? The Vidocq Experiment
A candid retrospective on building a Jakarta EE Core Profile and MicroProfile runtime from scratch. It covers architectural tradeoffs, JPMS-driven design changes, CDI event lessons, spec interpretation, dependency minimization, governance decisions, and what they would do differently with JDK 25 and hindsight.
talk.summaryAiDisclaimer
Antoine Sabot-DurandSCIAM
yann BlazartSCIAM
Sebastien BlancSCIAM
talkDetail.whenAndWhere
Monday, October 5, 13:30-16:30
TBA 6
talks.roomOccupancytalks.noOccupancyInfo
This conference started in 2001 as Javapolis, the same year J2EE's first specs were taking shape. Twenty-five years later, on this stage, we're closing that loop: implementing those specs with a 2026 AI coding agent as co-author, live.
Part 1 — Building Vidocq, then building something new, live. We'll walk through how Vidocq, our Jakarta EE Core Profile 11 and MicroProfile 7.1 runtime, came together: the choice to code against formal specifications rather than our own assumptions, and what changed once AI joined that process. Where AI-generated code passed our review but failed spec conformance. Where a spec's edge cases caught something the AI happily hallucinated past. The workflow we landed on to keep AI output honest against TCK-level requirements. Then we'll do it live: implementing a Jakarta EE or MicroProfile spec we haven't built yet, on stage, with an AI agent constrained by the spec text. Real live coding, backup video included in case it goes sideways.
Part 2 — Vidocq, from spec to CLI. A fast tour of what defines the runtime: JPMS enforced at unit granularity, Build Compatible Extensions, and the JDK 25 features (virtual threads, structured concurrency, scoped values) behind Vidocq's zero external dependencies. Then straight to the command line: we'll bootstrap and build a small app on Vidocq live, showing how those choices feel to work with, not just how they look in a diagram.
Two decades of spec work, one AI model, one CLI. Watch the gap close.
Part 1 — Building Vidocq, then building something new, live. We'll walk through how Vidocq, our Jakarta EE Core Profile 11 and MicroProfile 7.1 runtime, came together: the choice to code against formal specifications rather than our own assumptions, and what changed once AI joined that process. Where AI-generated code passed our review but failed spec conformance. Where a spec's edge cases caught something the AI happily hallucinated past. The workflow we landed on to keep AI output honest against TCK-level requirements. Then we'll do it live: implementing a Jakarta EE or MicroProfile spec we haven't built yet, on stage, with an AI agent constrained by the spec text. Real live coding, backup video included in case it goes sideways.
Part 2 — Vidocq, from spec to CLI. A fast tour of what defines the runtime: JPMS enforced at unit granularity, Build Compatible Extensions, and the JDK 25 features (virtual threads, structured concurrency, scoped values) behind Vidocq's zero external dependencies. Then straight to the command line: we'll bootstrap and build a small app on Vidocq live, showing how those choices feel to work with, not just how they look in a diagram.
Two decades of spec work, one AI model, one CLI. Watch the gap close.
Antoine Sabot-Durand
Antoine is a Java Champion and former spec lead for Jakarta EE CDI. He was part of the founding team behind Quarkus at Red Hat and worked on major MicroProfile specifications such as Fault Tolerance and Health Check. Today, Antoine is an architect and technical expert at SCIAM, where he leads complex projects largely built on the very technologies he helped design. He is also the co-creator of the Vidocq framework, a modern MicroProfile stack built on Java 25 and JPMS.
yann Blazart
Yann Blazart is a Senior Tech Lead and a passionate advocate of Jakarta EE and Quarkus. He has been in love with Java for 26 years and has been working with SCIAM since 2019. A long-time open-source contributor, Yann has taken part in projects such as WicketExtJs, TomEE, and JAX-WS Metro.
He is no stranger to the Devoxx stage: he ran two Hands-on Labs at Devoxx France (2012 and 2015) on Java EE, and more recently delivered a Hands-on Lab on Quarkus and Apache Ignite at Riviera Dev, Devoxx BE, and Devoxx MA in 2023. In 2024, he laid the foundations of what has since become langchain4j-cdi. In 2026, together with Antoine Sabot-Durand, he built Vidocq (https://vidocq.dev) — the first fully European MicroProfile 7.1 runtime.
He is no stranger to the Devoxx stage: he ran two Hands-on Labs at Devoxx France (2012 and 2015) on Java EE, and more recently delivered a Hands-on Lab on Quarkus and Apache Ignite at Riviera Dev, Devoxx BE, and Devoxx MA in 2023. In 2024, he laid the foundations of what has since become langchain4j-cdi. In 2026, together with Antoine Sabot-Durand, he built Vidocq (https://vidocq.dev) — the first fully European MicroProfile 7.1 runtime.