Rooms promise continuity, not just more agents
Nomadian’s most concrete promise is not the number of AI teammates; it is that work can continue with context. Hancom says users can keep separate Rooms for clients or projects, run several jobs at once, and review each agent’s status, progress, and tool use. Its Korean product page says agents remember a user’s work preferences within the same Room, carry them into later results, and leave an execution history that can be reviewed later. It names five roles—project lead, growth marketer, business writer, creative director, and customer-care manager—and lists integrations including Google Workspace, Slack, Notion, Canva, Shopify, Asana, and GitHub.
That is a more specific product idea than “AI employees”: a stateful workspace for recurring work. MTN reported that users could also create and assign agents for their own roles; Hancom’s current page lists standard roles but does not explain how custom ones would work.
Nomadian is still a plan, not a service people can evaluate. Hancom lists an open beta for December 15, 2026; its English page asks users to join a waitlist and warns that demo visuals and features may change. MTN reported a U.S. beta plan, paid subscriptions targeted for 2027, and pricing and distribution to be decided after beta feedback.
The wider stack is relevant—but not proof
Hancom says actions it considers hard to reverse—such as sending email, requesting an electronic signature, or making a payment—will pause for approval. MTN also reported that Nomadian uses Hancom’s Agentic OS as its underlying technology. Those are useful claims, but no public test shows whether an approval technically blocks an action at the connected app, rather than only pausing the interface.
Hancom’s separate Agentic OS page describes broader enterprise features: selecting or combining commercial and open-source models, connecting systems such as ERP and CRM, on-premises or closed-network deployment, access controls, guardrails, audit logs, and automatic stops for runaway loops. A different product, Orchestration Grid, claims runtime permission checks and protected credentials. These pages make Hancom’s wider platform strategy more concrete, but they do not say which of those controls Nomadian’s small-business SaaS actually inherits. MTN’s report of separate Agentic OS proof-of-concepts, including with Namyang Dairy, is likewise not evidence of a Nomadian customer deployment.
Memory turns privacy into a product question
A preference that carries from one task to the next has to be stored somewhere; an execution history does too. Hancom’s AI privacy policy says a separate policy may apply to an individual service. The reviewed policy lists AWS’s Seoul region for service-system operation and storage, and Google Analytics and Microsoft Clarity for website behavior analysis. It does not establish how Nomadian handles business prompts, connected-app content, remembered preferences, or run histories—or who can access and delete them.
The same gap applies to costs. Hancom says billing is usage-based and work automatically pauses at a user-set limit, but no public rate card or detailed metering rules are shown. It is not clear what the cap measures or what happens to a task already in progress when the limit is reached.
What the beta needs to demonstrate
A useful pilot should answer four questions:
- Can users inspect, correct, and delete what a Room remembers—and see how long run histories are retained?
- What can each connector read or change, and does approval block the action at the connector boundary?
- Do denied or expired approvals, retries, and failed handoffs leave a clear record without repeating risky actions?
- What exactly is metered, and does the spend limit stop work predictably, including in-flight jobs?
Persistent context and reviewable runs could make Nomadian more useful than a collection of chatbots. They also raise the bar: the beta has to show that memory, history, approvals, and spending limits are controllable—not just visible in a polished Room.