Jeff Ng is a founding engineer at Unblocked, building systems that give autonomous software agents the organizational knowledge engineers ordinarily provide themselves. His work addresses a consequential failure mode: agents can inspect code and operate tools while misunderstanding the incidents, conversations, and decisions behind a system’s design.
Ng studied at McGill University and works in the Vancouver area. At AI Engineer World’s Fair 2026, he demonstrated a Linear ticket-enrichment agent that investigated latency in a quality-assurance pipeline and recommended restoring asynchronous dispatch. The suggestion seemed technically sound, but the same configuration had already caused an outage. Because the agent lacked access to the subsequent Slack discussion and incident postmortem, it proposed repeating a known failure.
- Institutional context for autonomous agents. Supervised coding assistants benefit from engineers who explain previous decisions and catch mistakes. Background agents need another way to acquire that organizational memory before errors propagate through tickets, code reviews, and other systems.
- A permission-aware context engine. Ng brings together code, documentation, tickets, conversations, and incident history, then ranks and reconciles relevant information while respecting access controls. Agents receive synthesized, task-specific context instead of disconnected documents.
- Connectivity is not understanding. His conference demonstration distinguishes Model Context Protocol integrations from genuine contextual reasoning: connections to Slack, Linear, and GitHub expose raw information but do not resolve contradictions, establish relevance, or control context-window costs.
When Ng grounded the same agent in the missing postmortem and engineering discussion, its recommendation changed from recreating an outage to preventing one.
