Summary
Jeff Ng, founding engineer at Unblocked, opens by surveying how much agent deployment has been simplified by cloud infrastructure players like Cloudflare, Vercel, and AWS, combined with frameworks like Flo and Mastra — reducing what once took a team a quarter to build down to a few hundred lines of code. But the talk quickly pivots to what has not been solved: agents still fail confidently when they lack the organizational context that human engineers carry implicitly.
To illustrate the problem, Ng demos a Linear ticket enrichment agent built on Flo and Cloudflare that analyzes a real performance degradation issue and recommends re-enabling async dispatch in Unblocked’s QA pipeline. The recommendation is plausible — but wrong. An engineer had explicitly disabled async dispatch days earlier after it caused an outage, and that knowledge existed only in a Slack thread and a postmortem ticket the agent never saw. The agent had no way to know what it did not know.
Ng proposes a context engine — a system that synthesizes task-relevant information from scattered organizational sources (code, tickets, Slack, incident logs, design docs), respects access controls, resolves conflicts across datasets, and delivers synthesized context to agents without requiring a human in the loop. As agentic deployments scale beyond supervised use, he argues this layer becomes the critical missing infrastructure between trivially deployable agents and reliably autonomous ones. The talk is aimed at engineering teams beginning to hit the ceiling of code-and-ticket context alone.
📺 Source: AI Engineer · Published August 21, 2026
🏷️ Format: Keynote Launch
