An AI Future Without the Lock-In — Remy Guercio, Tailscale

This talk, presented by Remy Guercio from Tailscale, addresses the growing issue of AI lock-in and the need for demonstrable ROI in AI deployments. It breaks down an AI stack into four layers: models, data connectors (MCP), CLIs/APIs, and the interface/sandbox. The core contribution is the concept of an AI gateway, exemplified by Tailscale's Aperture, which acts as a central point for managing identity (via Tailscale network), budgets per model, and logging. Aperture functions as both an LLM and MCP gateway, consolidating authentication for various tools and providing a unified log for diagnostics, thereby enabling swappability of underlying AI models and services.