The energy on the second day of the summit in San Francisco remained incredibly high as the focus shifted from foundational infrastructure to workflow orchestration. If the first day was about ensuring autonomous systems have access to fast and accurate data, the second day was about managing the sheer variety of tools those systems use. Engineering teams are no longer relying on a single vendor or a single framework. They are deploying a massive combination of models and wrappers to handle specific tasks, which creates a highly fragmented environment. The keynotes addressed this fragmentation directly by introducing open standards for coordination and new methods for team collaboration.

Orchestrating Workflows with the Omnigent Meta-Harness

The most pressing challenge for developers right now is the isolation between different AI tools. To solve this coordination problem, Databricks announced the open source release of Omnigent, which functions as a meta-harness. A meta-harness is a coordination layer that sits directly above individual agent frameworks to provide a single unified interface. By treating each individual harness as an interchangeable component, Omnigent allows organizations to easily switch between models using a simple configuration change while maintaining their existing workflows. Beyond technical orchestration, it fundamentally changes how human teams interact with autonomous software by making live sessions fully collaborative. Engineers can share an active agent session via a simple web link, allowing teammates to join, review files, and steer the autonomous behavior together in real time.

Automating Security Operations with Panther

Another major announcement during the second day was Databricks’ intent to acquire Panther Labs to advance their vision for the security lakehouse. Legacy security information and event management systems, commonly called SIEMs, were simply not built to handle the speed and complexity of modern attacks. By bringing Panther’s artificial intelligence security operations center platform into the fold, engineering teams gain access to agentic workflows that can automatically triage alerts and gather context. This allows organizations to use swarms of agents for immediate threat investigation and response, directly on top of their unified data storage.

Managing the Soaring Costs of AI

During his keynote remarks, Databricks CEO Ali Ghodsi directly addressed a growing pain point for technical leaders, noting that agentic development is going to get extremely expensive as consumption increases. When autonomous systems are allowed to run continuously, they can easily consume massive amounts of computing resources in a very short period. To help organizations govern these expenses, Databricks highlighted new enhancements to the Unity AI Gateway. This tool provides contextual security policies and the ability to monitor costs and set strict budgets. Smart routing features also help teams automatically find the most cost-effective model to execute a specific task, ensuring that innovation does not lead to unmanageable invoices.

Enhancing Omnigent with LangGuard via Open Standards

While Omnigent provides excellent coordination and Unity AI Gateway offers strong baseline cost controls, enterprise security teams require an independent layer of authorization. This is where a Policy Decision Point becomes absolutely critical. A Policy Decision Point, or PDP, is a centralized security engine that evaluates every single action an application attempts to take and compares it against a strict set of corporate rules before granting permission. Relying on an agent to police its own behavior is a significant risk, so the PDP must operate completely independently from the agent itself.Luckily, Omnigent has an extensible policy framework available to add support for external PDPs easily.

LangGuard has extended Omnigent to add support for the burgeoning MCP Interceptor specification from the Agentic AI Foundation, who owns the MCP standard. This will let you connect any PDP that supports the standard (such as LangGuard Arbiter) to Omnigent. Once connected, LangGuard Arbiter acts as an exact deterministic policy engine for multi-agent workflows in Omnigent. As organizations deploy tools like Omnigent to coordinate massive fleets of agents, the LangGuard platform evaluates every intended tool call across the entire ecosystem. Before an agent can call a tool, the Arbiter cross-references the request against established segregation of duties frameworks and compliance standards. If the action violates a security policy, LangGuard automatically pauses the execution and requests human review. This ensures that organizations can scale their autonomous operations rapidly while maintaining absolute confidence in their security posture.

Conclusion

The second day of the summit provided a clear roadmap for the future of enterprise software development. By open sourcing tools like Omnigent and expanding the security lakehouse, the industry is moving away from fragmented environments and toward highly interoperable ecosystems. When these powerful orchestration layers are paired with the deterministic security controls provided by LangGuard, businesses can finally move their most ambitious autonomous projects out of the testing phase and into secure production environments.