More Context Hurts AI Agents; Curation Beats Dumping

Adding more context to AI agents often degrades output due to bloat, dilution, conflict, and latency, according to engineers at Stripe and CodeRabbit. Teams that succeed shift from dumping data to curating what agents see via a context engine that unifies, retrieves, ranks, resolves, governs, and scopes information. The real challenge is that undocumented intent and trade-offs—the submerged part of the iceberg—remain inaccessible to agents.