Technology
November 12, 2026 1001 words

Enterprise AI Agents Grow Up: Databricks Always-On Genie, Kimi Hosted Agents and More

Hams AI Tech
Professional Content Team
Expert in AI Prompts & Professional Tools

For most of the past two years, enterprise AI agents were experiments: impressive demos that struggled in production. In November 2026, several vendors launched platforms designed for real, long-running enterprise work. Databricks turned Genie into an always-on AI coworker, Moonshot AI launched Kimi Hosted Agents as a managed harness for multi-agent workloads, and ElevenLabs added memory and routines to its voice agents. These launches share a common goal: making agents reliable, governed and safe enough for serious business processes.

Databricks: The Always-On Genie

On November 11, Databricks announced a new version of Genie that works as an autonomous AI coworker for long-running enterprise jobs. Key capabilities include:

    • Running on schedules or triggers, not only when someone asks a question.
    • Following up on missing inputs and asking for approval where needed.
    • Availability in Genie One, email, Slack and Microsoft Teams.

Databricks also introduced several supporting features:

    • Genie Ontology: business context built by agents and governed in Unity Catalog, so the agent understands what terms like revenue, customer or active account mean in your organization.
    • Genie Search: search across the lakehouse and apps such as Google Drive, Jira and SharePoint.
    • Shareable artifacts for distributing results.
    • MCP connectors and plugins for ChatGPT, Claude, Copilot and Grok Bot.
    • Unity Gateway spend and access controls.

The always-on Genie is available to Databricks enterprise customers only.

Why Business Context Matters

The biggest reason enterprise agents fail is not intelligence but context. An agent that does not know your definitions, data sources and approval rules will produce plausible but wrong answers. Genie Ontology tackles this directly by building and governing business context. Anthropic's certified queries in Claude Data solve a similar problem: making sure everyone gets the same answer to the same question.

Kimi Hosted Agents: Harness as a Service

On November 5, Moonshot AI launched Kimi Hosted Agents, a managed service for running long-horizon and multi-agent workloads without building custom infrastructure. It includes:

    • Isolated cloud sandboxes.
    • Network controls and stoppable sessions.
    • Audit logs.
    • More than 140 skills and 30 plugins.
    • Eight finance agent templates.

Pricing is usage-based, combining sandbox hours, model tokens and plugin calls. Enterprise-verified users can add Kimi Inference Hook for safety checks before inference. Our Kimi K3.1 Pro article covers the model behind it.

ElevenAgents: Memory and Routines for Voice

At its New York summit in November, ElevenLabs announced Memory and Routines for ElevenAgents, its voice agent platform, along with an Interaction Engine preview that routes each part of a conversation to a cheaper or frontier model. Voice agents that remember customers and follow routines can handle more complex service calls. See our voice AI roundup for details.

What Enterprise-Grade Agents Need

Across these launches, a consistent set of requirements emerges:

    • Governance: clear rules about what data agents may access and what actions they may take.
    • Isolation: sandboxes that prevent agents from affecting systems they should not touch.
    • Auditability: logs that show what an agent did and why.
    • Approvals: human sign-off for consequential actions.
    • Cost controls: spending limits and monitoring.
    • Context: shared definitions and trusted data sources.

Any organization evaluating agent platforms should use this list as a checklist.

How Professional Firms Can Apply These Ideas

Large enterprises will adopt platforms like Databricks Genie directly. Smaller professional firms can apply the same principles with lighter tools.

Accounting and Finance

Finance teams can build always-on agents that reconcile accounts, flag anomalies and prepare monthly reports. Our multi-entity consolidation agent, expense fraud detector and IFRS and GAAP compliance checker follow the same scheduled, approval-based model at a smaller scale.

Law Firms

Agents can monitor dockets, prepare case summaries and track deadlines. See the case summary auto-builder and the paralegal task automation suite.

Operations and Management

Managing several agents requires a control center. The AI Agent Management System Notion template provides a dashboard for tracking agents, tasks, outputs and schedules. Browse more in command center resources and management systems.

Choosing a Platform

    • If your data lives in Databricks, the always-on Genie is a natural fit.
    • If you need a managed environment for custom multi-agent work, consider hosted harnesses such as Kimi Hosted Agents.
    • If your agents interact with customers by voice, look at platforms like ElevenAgents.
    • If you are a smaller firm, start with workflow tools such as n8n and a clear governance policy, then grow.

Common Pitfalls to Avoid

Organizations that struggle with agents usually make one of a few mistakes. The first is starting with a task that is too broad, such as asking an agent to manage all customer communications, rather than a narrow, measurable job. The second is skipping the context step, which leaves the agent guessing at definitions and data sources. The third is granting too many permissions too early, so a small error can have large consequences. The fourth is failing to assign an owner: every agent should have a named person responsible for its configuration, performance and incidents. Avoiding these mistakes matters more than choosing the perfect platform.

Frequently Asked Questions

Who can use the always-on Genie?

It is available to Databricks enterprise customers.

What is Genie Ontology?

Business context built by agents and governed in Unity Catalog, helping Genie understand organization-specific terms and data.

How is Kimi Hosted Agents priced?

Usage-based, combining sandbox hours, model tokens and plugin calls.

Do enterprise agents act without approval?

The platforms described here support approval steps for consequential actions, and organizations should configure them carefully.

The Bottom Line

Enterprise agents are moving from experiments to governed, always-on coworkers, and the tooling to run them safely is finally arriving for organizations of every size. The winners will be organizations that pair capable platforms with clear context, approvals and audit trails. Explore our AI agents section and agentic AI resources to design agents that work in the real world.

Related AI News

Topics & Keywords
enterprise AI agents Databricks Genie Genie Ontology Kimi Hosted Agents ElevenAgents AI coworker Unity Catalog