Superagentic AI is proud to announce the Agent Engineering Conference.  For the past few months, I’ve been quietly working on something, since out last event exploded to registration to 1900+  Over the last few years, we’ve watched AI evolve in phases. First came the model breakthroughs. Then prompt engineering became the dominant craft. Then we began wiring tools together, building simple chains, and calling them agents. But as teams started deploying agents into production systems, something changed.

The problem was no longer about writing better prompts. It was no longer about model capability alone. It became about how these systems behave over time. How they fail. How they recover. How they scale. How they are evaluated. How they are orchestrated. And in that shift, a new discipline started to take shape.

Agent Engineering.

We moved from prompt engineering to context engineering. From harness design to evaluation frameworks. From static scripts to systems that manage memory, tools, feedback loops, and human oversight. Every layer added complexity. Every layer introduced non-determinism. Every layer demanded more rigor. All roads began to lead toward something bigger than prompts or frameworks. They led toward engineering systems around intelligence.

Agent Engineering is the discipline of specifying, designing, evaluating, and orchestrating non-deterministic AI systems in production. It treats agents not as demos or features, but as systems that must be observable, reliable, and maintainable over time. As foundation models become increasingly capable and increasingly commoditized, the real differentiator shifts. It shifts toward architecture. Toward orchestration. Toward evaluation. Toward memory management. Toward operational guardrails. It shifts toward engineering.

Yet if you look at the conference landscape today, most AI events remain broad. Many are marketing-heavy.  What’s missing is a highly technical, curated room focused purely on this emerging discipline.

A room where engineers can walk through real systems. Where speakers show logs, IDEs, traces, and evaluation dashboards instead of slide decks. Where non-determinism is discussed honestly. Where failure is treated as signal. Where production metrics matter more than product messaging.

That is why we are launching AgentEng 2026.

The Agent Engineering Conference.

The first edition will take place in London in Summer 2026, rooted in the 2,000+ member London Agentic AI community that has grown around practical, production-oriented conversations. A larger edition is planned for San Francisco later in the year, extending the bridge between London and Silicon Valley.

AgentEng will be single-track. Focused. Intentional. Demo-first. Highly technical. It will center on the real disciplines that underpin agentic systems: orchestration, evaluation, memory, context construction, harness design, and multi-agent coordination. It will prioritize engineers who are actively building and operating these systems in production, not just discussing them in theory.Throughout 2025, the term “Agent Engineering” began surfacing independently across practitioner communities, research labs, and platform builders. The language may still be forming, but the need is clear. The systems are becoming more complex. The expectations are rising. The engineering standards must evolve with them. AgentEng exists to bring those parallel conversations into one place, focused on practice rather than promotion.

The call for speakers is now open. Founding partners are invited. Tickets will open soon.

If you are building and operating real agentic AI systems, this is your room.

https://agentengineering.world

The next phase of AI will not be defined solely by model capability. It will be defined by how intelligently we engineer the systems around it. That work deserves its own discipline. And its own conference.