AI agents are only as valuable as the experiences they create. The enterprises getting this right aren’t just deploying more agents – they’re deploying agents people can actually rely on and trust.
At Connect: AI New York, we were joined by hundreds of business and technology leaders and MuleSoft community leaders from across industries to tackle the question that’s top of mind for every IT team right now: how do you accelerate the move to the agentic enterprise without losing visibility, control, or confidence along the way?
Every hyper-scaler, data lake, and application is evolving into an AI platform. The result is a new era of bring-your-own-AI. Without the right controls in place, individual teams plugging in their own services aren’t just adding capability. They’re adding risk.
We spotlighted the releases tackling both sides of that challenge: tighter governance and security controls, paired with new partnerships and integrations that extend the ecosystem’s reach across new platforms and geographies, and the tools your teams already work in every day.
Omni Gateway: Governance built for speed, everywhere your AI and APIs already run
Speed and safety usually pull against each other. Add more oversight and teams can slow down; let them move fast and oversight slips. For most platform teams this has been the defining tension of the agentic shift: agents are reaching production faster than the infrastructure underneath them can govern, and every new control feels like a tax on momentum.
The enterprises getting AI to production fastest have figured out the answer isn’t more coverage, it’s consistency. Extending the governance disciplines that already work across the full stack, rather than rebuilding them surface by surface. Omni Gateway is built on that premise.
By making the gateway the single place where access is granted, spend is capped, and policy is enforced, the same controls that keep agents in check are what let teams move. A budget cap lets a team run hard without anyone watching the bill. A policy that holds across every gateway is what makes it safe to say yes to the next use case.
With these recent and upcoming releases, governance lives in the infrastructure instead of sitting on top of it, which means confidence stops being something teams trade away for speed; rather, it’s the reason they can push harder.
Federated Gateway Management: Govern every gateway from one place | GA now
As gateways multiply across teams, regions, and clouds, visibility fragments and policy drifts. Federated Gateway Management brings those distributed gateways under a single governance and management layer, so spend caps, access rules, and policy apply consistently across all your gateways.
Universal Policies: One policy enforced everywhere | GA Q3 2026
Setting rules agent by agent is how gaps appear. Universal Policies lets you define a rule once and have it hold across every agent and gateway in your estate, so a single change propagates everywhere it applies without reconfiguring each endpoint one at a time.
Cap what every agent can spend | GA Q3 2026
The cost of running agents can compound quietly, and the bill usually arrives before the visibility does. New spend controls in Omni Gateway put a ceiling on that. Each team or agent routes through its own key with a built-in budget cap, its own usage analytics, and rules for which models it’s allowed to touch.
Think of it like a Model Wallet per team rather than a shared account no one can trace – spend becomes attributable and capped before it turns into a surprise. The real provider credential sits behind the key rather than inside application code, which means users can rotate or restrict access without redeploying.
Agent Kill Switch: Stop any agent, at any scale, in seconds | GA Q3 2026
When an agent starts doing something it shouldn’t, the cost of waiting is measured in seconds. Agent Kill Switch affords businesses a stop that matches the size of the problem, with three tiers of defense: pause a single session with its state intact, freeze one agent type, or halt the entire fleet under a deny-all policy.
Enforcement is driven by the control plane, not left to the agent to honor, so a misbehaving agent can’t ignore an instruction to stop. The result is the confidence to put agents into production knowing that if something goes wrong, there’s a mechanism for an appropriately measured response.
Secret References: Keep credentials where they belong | GA Q3 2026
Connecting agents to enterprise credentials shouldn’t mean copying secrets into yet another system. Secret References point directly at the credentials already living in your existing AWS, Azure, or HashiCorp vaults, fetched only when needed and never stored. Security teams keep managing rotation exactly where it happens today, and every agent picks up the change automatically. The keys never leave the trusted vault, which means extending agents across an environment doesn’t widen their security exposure.
Agent Fabric: Manage, govern, and optimize your multi-vendor AI landscape
The agentic enterprise needs more than agents that can act. It needs a foundation that knows what every agent is built on, what data it can trust, and how it behaves once it’s live. The latest Agent Fabric releases tighten that foundation across the full lifecycle, from how orchestration is authored to the data agents draw on to act.
Enhanced Agent Broker: Balance determinism and agentic agility | GA July 2026
Agent Broker introduces three powerful capabilities designed to deliver better predictability, streamlined authoring, and expanded model flexibility for the enterprise. This update incorporates Agent Script, an expression language that gives you a hybrid approach combining deterministic workflows with agentic reasoning.
Developers can see these hybrid paths with a graph-based visualization and use MuleSoft Vibes to help select assets and scaffold the orchestration. Additionally, we’re expanding model support to Salesforce-managed models, alongside OpenAI and Gemini, enabling you to consolidate usage. That’s not all. We’re also debuting a local deploy and run capability that slashes testing cycles from minutes to seconds through real-time, side-by-side behavioral validation.
A2A Bridge: Bridging agent communication | GA July 2026
Building on the success of MCP Bridge, MuleSoft is introducing A2A Bridge to help IT leaders coordinate disparate AI agents built on different platforms and standards by making them A2A-compliant. GAing in July is supported for Salesforce Agentforce and Microsoft Copilot Studio with more platforms coming soon.
By dynamically handling protocol translation, ID mapping, and task lifecycle state management, the A2A Bridge allows organizations to bring their existing, non-A2A compliant agents under a single, centralized governance and orchestration umbrella without requiring developers to rewrite any underlying code.
Agent Fabric Context Catalog: Manage AI and data assets from a single control plane | GA July 2026
To deliver the trusted data foundation every AI agent needs, we’re bringing your enterprise data catalog directly into Agent Fabric. This new Agent Fabric Context Catalog dynamically unifies your Informatica data assets alongside all of your agents into a single, living lineage graph. For the first time, you can see, govern, and trust every vetted AI agent, every enterprise data asset, and the governance policies that control how agents are built, deployed, and act – with full lineage at the point of integration.
Expanding the ecosystem’s reach
The agentic enterprise doesn’t live in one place – and neither does MuleSoft. Here’s how the ecosystem’s reach keeps growing.
MuleSoft and Akamai Strengthen API and AI Governance with Expanded Discovery | GA July 2026
As enterprises embrace AI-driven applications and agentic architectures, securing and governing the APIs and MCP servers that connect them has become increasingly critical. MuleSoft and Akamai announced an expanded partnership that brings Akamai API Security insights directly into MuleSoft, helping customers discover unmanaged assets, assess risk, and apply consistent governance across APIs, agents, and MCP servers.
The partnership extends MuleSoft’s existing discovery capabilities by leveraging Akamai’s traffic analysis to identify previously unknown APIs and MCP servers, even in environments where MuleSoft has not deployed dedicated scanners. When Akamai detects new assets through runtime traffic, code repositories, or documentation sources, they can be automatically added to MuleSoft’s Agent Registry, creating a more complete and continuously updated inventory of enterprise assets.
By combining MuleSoft’s governance and cataloging capabilities with Akamai’s API discovery, risk scoring, and behavioral analysis, customers gain a unified view of APIs, agents, and MCP assets across their organization. Together, MuleSoft and Akamai make it easier to build a comprehensive catalog of enterprise assets, reduce security blind spots, strengthen governance, and confidently scale AI and agentic innovation across hybrid and multi-cloud environments.
The MuleSoft integration for Akamai API Security will be available to joint customers in July 2026.
MuleSoft on Hyperforce in India | GA now
MuleSoft is now generally available in India through Hyperforce, Salesforce’s public cloud infrastructure built for organizations that need strict data residency and compliance controls. Hyperforce co-locates the control plane and runtime plane within India, giving regulated industries (i.e. financial services, capital markets, healthcare, and the public sector) a way to run MuleSoft with data stored locally, without sacrificing the scale or security of the broader platform.
The bigger opportunity here goes beyond compliance. With Agent Fabric, Omni Gateway, and MuleSoft Vibes now available in the India region, organizations get a locally-hosted foundation for building and governing AI agents from day one – meaning the same auditability and data residency guarantees that regulated industries require now extend to agentic workflows, not just integrations.
India joins the US, EU, Canada, and Japan as part of MuleSoft’s broader Hyperforce expansion, putting compliant agentic infrastructure within reach for one of the world’s fastest-growing regulated markets.
Agent Scanners reach Databricks, Snowflake, Langsmith, and Claude | GA Now
As agentic development becomes even more decentralized and teams build across their platforms of choice, the need for a single, proactive view of all AI services running across your organization has never been more paramount. That’s why we’ve expanded Agent Scanners to support even more leading platforms.
Rather than managing disconnected and unvetted AI resources manually, organizations can now use Agent Scanners to automatically find and register agents from Databricks, Snowflake, Langsmith, and Claude – in addition to previously supported platforms like Salesforce Agentforce, Amazon Bedrock and Agentcore, Microsoft Copilot Studio and Azure AI Foundry, and Google Vertex AI. The recently announced Agentic Resource Discovery (ARD) specification will make these agents even more securely discoverable across the open agentic web.
MuleSoft Agent for Slack: Manage your ecosystem where your team already works | GA now
The MuleSoft Agent for Slack brings your entire agent and API ecosystem into the place where teams already work.
Ask questions in natural language:
- “Which agents are running outside policy?”
- “Which APIs are failing conformance checks?”
Or anything you can think of, and get permission-scoped answers instantly, without context-switching into Anypoint Platform. Real-time alerts for governance violations, performance issues, and deployment failures route directly to the right channels and DMs, and the agent doesn’t stop at notifying you: it flags violations and generates remediation steps on the spot.
For teams using Slackbot, connecting your MuleSoft MCP server is now generally available and brings the same visibility and governance into that experience as well.
The agentic enterprise, built with trust, powered by MuleSoft
Every conversation at Connect:AI New York this week came back to the same idea: the teams that move with confidence aren’t the ones who ignored the guardrails – they’re the ones who got them right.
When agents are governed, costs are capped, credentials are secure, and your ecosystem is visible from wherever your team already works, something shifts. The hesitation goes away. New experiences go into production. That’s what this infrastructure is built for: not to slow the agentic enterprise down, but to give it somewhere solid to stand on.
Miss Connect:AI New York? Catch up on everything that happened – sessions, announcements, and more – in the full recap.