From Copilots to Agentic SE Teammates - Engineering the SE 3.0 Era with AI-Native Workflows

Explore the evolution from AI copilots to autonomous software engineering teammates in this conference talk that introduces the SE 3.0 paradigm. Discover why current foundation model-powered copilots fall short due to developer overload, fragmented workflows, and lack of agentic behavior, then learn about the comprehensive SE 3.0 Agentic Stack including Teammate.next for adaptive AI partners, IDE.next for intent-first development environments, Compiler.next for multi-objective code generation, and Runtime.next for SLA-aware execution environments. Examine findings from AIDev, the first large-scale dataset capturing AI-agent-generated pull requests across 61,000+ repositories, which reveals both the potential and current limitations of agentic contributions including acceptance, coordination, and complexity challenges. Understand why real-world agentic data exposes gaps that synthetic benchmarks like SWE-bench miss, and explore research and engineering opportunities for benchmarking agent trust, modeling collaboration, and embedding governance into Agentic SE workflows. Gain insights into building the metrics, tooling, and safeguards necessary to fully realize the potential of Agentic Software Engineering 3.0 in practical applications.