Joseph Wang is the co-founder of Emulated, which builds training and evaluation environments for autonomous software engineers; he served as its chief executive at AI Engineer World’s Fair 2026. His focus is the gap between agents that write convincing code and those capable of operating live production systems safely.
Wang and his co-founder, Sid, bring backgrounds in network infrastructure, distributed databases, and sandbox infrastructure. Their approach to infrastructure-aware agent autonomy incorporates operational details conventional coding tasks exclude: customer context, incident histories, network failures, cloud provisioning, staged deployments, monitoring, and ongoing live traffic.
- Software Development Automation Benchmark: Published in April 2026, Emulated’s 80-task production-engineering benchmark evaluates infrastructure debugging, migrations, deployment pipelines, incident response, and distributed systems. Tasks run for up to 12 hours, with scoring based on working systems, operational judgment, and engineering quality.
- Cloudboxes and multi-node sandboxes: Emulated builds isolated environments spanning real cloud resources and distributed infrastructure, exposing agents to conditions single-container simulations cannot fully reproduce. Its scenarios include migrating databases without losing active writes, reversing deployments that degrade live traffic, and provisioning Kubernetes clusters for distributed training.
- Production outcomes over plausible code: Wang treats successful autonomy as an operational result: services recover, migrations preserve data, monitoring detects regressions, and agents verify their changes under realistic load. Infrastructure provides an initial proving ground for his broader ambition: environments sophisticated enough to prepare agents for responsibilities across entire software organizations.
