The industry's obsession with token optimisation—squeezing maximum performance from LLM context windows through prompt engineering and data compression—has reached its practical ceiling. What's emerging instead is a fundamental shift toward agentic memory systems that allow AI agents to retain, retrieve, and reason over customer interaction history across sessions, rather than relying on cramming everything into a single request. This transition reflects a maturation in how organisations approach AI-assisted customer service: moving away from stateless, single-turn interactions toward persistent, contextually-aware agents that can genuinely understand customer journeys. For CX teams already invested in Zendesk or Salesforce integrations, this means the value proposition of your AI layer is no longer determined by how efficiently you compress prompts, but by how effectively your system remembers and acts on what it knows about each customer.
The practical implications are substantial. Token-maxxing created a false economy where teams optimised for cost per interaction rather than outcome per relationship. Agentic memory inverts this: systems that can access structured customer history—previous issues, preferences, failed resolutions, sentiment patterns—will outperform those relying on real-time context alone, even if they consume more tokens overall. This raises a critical question for support leaders: are your current platforms architected to support persistent agent memory, or are you locked into stateless query-response patterns? Vendors like Salesforce with Agentforce have already begun embedding memory capabilities, but many mid-market CX stacks haven't. The gap between platforms that can maintain agent state across customer interactions and those that cannot will likely become the primary differentiator in the next 18 months, making this less a technical curiosity and more an urgent infrastructure decision.
Presented by MongoDB We have been building databases as an industry for roughly 60 years. We have been building AI agents, in the form most people mean when they say the word today, for about 18 months.Sit with that ratio for a second, because it explains almost everything about the state of agentic