Why Your AI Assistant Has Dementia: The $72B Identity Crisis (medium.com)

In September 2025 a broad, cross‑platform crisis surfaced: AI assistants consistently lose session context, forcing users to re‑teach agents every time they switch tools or continue long workflows. Painful, documented failures—from developers losing 225 lines of code to users reporting 40KB+ of instructions vanishing after a few prompts—have coalesced into what practitioners call “Context Degradation Syndrome” (CDS). Empirical evidence compounds the outrage: METR’s study found developers using AI actually took 19% longer, even though they felt 20% faster. Power users increasingly disable built‑in memory because it saves noisy, irrelevant details that “pollute” the context window, and multi‑agent pipelines collapse when each agent starts blind. Technically, the problem stems from architecture and operational design: context rot (more history can worsen output), limited and brittle context windows, model updates that erase state, and centralized vendor silos that trap fragmented operational knowledge. The result is wasted tokens, duplicated work, compliance and data‑sovereignty risks, and broken multi‑step automation. The community consensus is clear: without a portable, persistent AI identity and robust memory infrastructure—one that spans sessions, models, and vendors—enterprise AI will keep hemorrhaging productivity. Whoever builds that interoperable identity layer first could capture a market estimated at $72B; until then, AI remains powerful but forgetful.