Blog
Notes on context for AI
Guides, comparisons and engineering notes on giving AI agents shared, governed context across the software lifecycle.
InsightsIs MCP enough for shared context between AI agents?No. In specification revision 2025-11-25, an MCP Resource declares nine fields and not one of them records which source is authoritative, which version is current, or which of two contradictory resources wins. A ride-sharing marketplace found out what that absence costs when an agent filtered on a valid word and returned nothing.Jul 13, 2026 · 12 min readInsightsHow to accelerate with AI when your company is starting lateStarting late is an advantage only if you skip stages: organize data, processes, and communication first, build a culture of working with AI, then adopt only AI that compounds. Late starters who skip the foundation replay the industry's mistakes in order and buy point acceleration that never reaches delivery.Jul 11, 2026 · 11 min readInsightsWhy is AI making your software delivery slower?Because AI accelerates the local work of producing code, specs, and tickets, while the work that gates delivery, agreeing on what is true, still runs at human speed. The fix is reconciliation at machine speed, not a smarter model.Jul 9, 2026 · 9 min readGuidesWhat is a context layer for AI agents?A context layer is a governed, shared source of truth that every AI agent reads before acting, so tools stay coherent instead of each guessing from a slice of the work.Jul 7, 2026 · 2 min readComparisonsMCP vs RAG: how should AI agents get context?RAG finds text similar to a query; MCP is the interface an agent uses to ask a source for context. They solve different problems, and neither one decides which source is authoritative for a fact. That job belongs to a governed context layer behind MCP, with retrieval running inside it.Jul 6, 2026 · 2 min read