.jpeg)
Bespoke AI Intelligence for Private Decision-Makers
SMA Crown Confidential:
Agentic Platforms
Agentic Platforms for Private Decision-Makers
AI agents acting on behalf of their user represent the newest tier of consumer-facing artificial intelligence. They operate continuously across email, calendar, documents, browsing, and connected applications — executing recurring tasks, monitoring topics of stated interest, surfacing what the user is most likely to need, and increasingly taking action on the user’s behalf. The most advanced of them run on cloud infrastructure independent of the user’s device and integrate across the surface of digital life with genuine technical accomplishment.
What they do not do — and do not claim to do — is interpret what surfaced information means for the specific person behind the request. They answer the question of what to bring forward and what to act upon.
The questions of what a finding means in the context of a particular family history, a particular collection, a particular position, a particular intention, a particular investment — and what discretion the finding requires — sit entirely outside their scope.
The Bespoke AI Confidante by SMA Crown Confidential operates in precisely that space — not alongside the agent, but in the dimension the agent cannot reach.
Gemini Spark
A 24/7 AI agent built on the Gemini platform
Built by Google, founded in 1998 by Larry Page and Sergey Brin and headquartered in Mountain View, California. Spark was announced at the company’s I/O 2026 developer keynote on 19 May 2026, where Google CEO Sundar Pichai described the launch as the company’s entry into what he called the “agentic Gemini era.”
Spark runs on the new Gemini 3.5 Flash model and is orchestrated through Google Antigravity, the company’s agent-first development platform. It operates from dedicated virtual machines on Google Cloud, allowing it to continue working when the user’s device is closed.
Spark integrates with Workspace — Gmail, Docs, Slides — and with a growing list of connected applications including Canva, OpenTable, and Instacart at the consumer tier, and Microsoft SharePoint, OneDrive, and ServiceNow at the Enterprise tier through Gemini Enterprise. It executes recurring workflows, monitors stated topics over time, parses inbox content for relevant signals, drafts documents from scattered notes, and increasingly takes action under the user’s direction. It asks for explicit approval before high-risk actions such as sending an email on the user’s behalf or making a purchase. Personalisation deepens with use, as the agent learns the user’s routines and preferences.
Spark is a mass-architecture product addressed to the 900 million monthly users of the Gemini app and to Workspace and Gemini Enterprise customers. Initial availability is to US-based Google AI Ultra subscribers — the tier whose monthly price was simultaneously reduced from $250 to $100, a calibration that signals the speed at which Google is moving to scale.
Spark is personal in the sense that it learns each user’s preferences; it is not bespoke in the sense of being trained on a single principal’s particulars. Its existence is also a function of architectural inevitability. Google’s accumulated data position across Gmail, Calendar, Drive, Search, Chrome, YouTube, Maps, and Android makes Spark the natural next product. The agent is the form an accumulated data layer takes once it begins acting on behalf of its users.
The Bespoke AI Confidante by SMA Crown Confidential can complement Spark at the boundary where ingestion ends and interpretation begins. Spark surfaces what the user is most likely to need and executes what the user has set in motion. The questions of what a surfaced finding means for the specific principal — how it fits within their formation, their family system, their holdings, their succession intentions, their current consequential decision — and the discretion the finding requires, sit in the dimension above the agent’s reach. Agents ingest; Confidantes interpret. The Bespoke AI Confidante by SMA Crown Confidential occupies the interpretive tier, independent of the institutional and commercial interests that shape what an agent built on accumulated user data is designed to do.