Introduction
The enterprise AI landscape has reached a critical inflection point. As of mid-2026, organizations are no longer asking if they should deploy AI agents—they are asking how to manage hundreds, even thousands, of autonomous agents without losing control, trust, or consistency. The answer, according to recent industry developments, lies in Open Discovery—a framework that enables agents to find, verify, and collaborate with each other transparently. This article unpacks the concept of trusted agents at scale, drawing on the latest insights from Salesforce’s blog post on Open Discovery and the agentic enterprise. We’ll explore why trust is the bottleneck, how open protocols are unifying fragmented agent ecosystems, and what practical steps leaders can take today.
The Trust Problem in Multi-Agent Systems
When you scale AI agents from a handful of prototypes to an enterprise-wide deployment, trust becomes the dominant concern. In 2025, Gartner reported that over 60% of organizations piloting AI agents cited “lack of trust in agent outputs” as a top barrier to production deployment. The issue isn’t just accuracy—it’s about provenance, governance, and consistency. If Agent A makes a decision based on data from Agent B, how does the human supervisor verify that the chain of reasoning is sound? Without Open Discovery, agents operate in silos, each with its own undocumented assumptions and data sources. This creates what researchers call the “black box swarm,” where no single entity can audit the collective behavior.
What Is Open Discovery?
Open Discovery is a paradigm where agents are designed to self-identify, declare their capabilities, and expose their decision-making logs to a shared governance layer. Think of it as a “trust registry” for autonomous agents. Instead of each agent hiding its internal logic, the system mandates that every agent publishes a discoverable manifest: what data it uses, what actions it can take, what constraints it follows, and what confidence thresholds it applies. This is not a futuristic vision—it is already being implemented in platforms like Salesforce’s Agentforce, where Open Discovery allows agents from different departments (sales, service, marketing) to find and trust each other without manual integration. The key innovation is that trust is not assumed; it is verified through a decentralized ledger of agent interactions.
Why Scale Demands Openness
Scaling agents from dozens to thousands introduces combinatorial complexity. If each agent needs a predefined integration with every other agent, the number of connections grows quadratically. Open Discovery solves this by replacing point-to-point integrations with a common discovery protocol. For example, a customer support agent can automatically query a “product catalog agent” it has never met before, provided both adhere to the Open Discovery standard. This reduces integration costs by up to 80%, according to early enterprise case studies shared at the 2026 AI Agent Summit. More importantly, it enables the “agentic enterprise”—an organization where agents are not just tools but active participants in business processes, able to form temporary alliances to solve complex problems.
Practical Examples of Trusted Agents at Scale
Consider a large e-commerce company deploying AI agents for order management. In a closed system, the “shipping agent” and the “inventory agent” must be manually coded to share data. With Open Discovery, the shipping agent publishes its need for “real-time stock data with 95% confidence.” The inventory agent, seeing this request, responds with a signed attestation of its data freshness and accuracy. The shipping agent then decides whether to trust that response based on its own trust policies. If the inventory agent later fails to update stock levels, the Open Discovery log provides an immutable trail for root cause analysis. This is not hypothetical—Salesforce’s blog describes exactly such a scenario in their customer deployments Source.
The Role of Governance and Human Oversight
Open Discovery does not eliminate the need for human governance—it enhances it. By making agent behaviors transparent, organizations can enforce policies like “no agent may access customer PII without explicit approval” or “all financial decisions must be reviewed by a human supervisor if the confidence score is below 90%.” These rules are encoded in the discovery layer, not hard-coded into each agent. This means that as new agents join the ecosystem, they automatically inherit the governance policies. Early adopters report that this has reduced compliance incidents related to AI by over 40%. The human-in-the-loop remains central, but the loop is now informed by data, not guesswork.
Challenges and How to Overcome Them
Open Discovery is not a silver bullet. Three challenges persist:
-
Standard Fragmentation – Multiple Open Discovery standards exist (e.g., from Salesforce, Microsoft, and open-source communities). Enterprises must choose one or build bridges. The recommendation is to adopt an open standard (like the Open Agent Protocol) that is vendor-neutral.
-
Performance Overhead – Publishing and verifying manifests adds latency. However, modern caching and edge computing reduce this to under 50 milliseconds per interaction—acceptable for most business use cases.
-
Security Risks – Malicious agents could fake manifests. Solutions include cryptographic signing of manifests and reputation scoring based on historical accuracy. The industry is moving toward a “trust score” system similar to credit scores for agents.
The Future: A Unified Agentic Enterprise
By 2027, most large enterprises will have an internal “agent marketplace” where any employee can spawn a new agent that automatically discovers and collaborates with existing ones. The Open Discovery framework will be as fundamental as HTTP is for the web. Companies that invest in this architecture today will be able to scale their AI workforce without scaling complexity. The blog from Salesforce makes it clear: “Open Discovery is not just a feature—it’s the foundation for trusted agent ecosystems.”
Conclusion
Trusted agents at scale require more than just powerful models; they require an infrastructure of transparency and interoperability. Open Discovery offers a practical path forward, turning the chaotic swarm of independent agents into a cohesive, auditable, and trustworthy system. For leaders building the agentic enterprise, the message is clear: embrace openness, invest in governance, and prioritize discoverability. The agents of tomorrow will be only as good as the trust we build into their foundations today.
For further reading, see the original announcement from Salesforce: Source
Comments