Yesterday's keynote at Boomi World 2026 in Chicago covered the strategic frame: data activation as infrastructure, the Lunar.dev acquisition, Red Hat collaboration, and Boomi Connect as the governed layer between enterprise systems and AI tools. I wrote about that here. Day 2 was the product layer and the customer evidence. Chief Product and Technology Officer Ed Macosky, Senior Vice President of Product Mani Gill, Chief Technology Officer Matt McClure, and Ann Maya, Global Head of Strategic Projects, moved through what that infrastructure actually does when it runs in production. The customer and partner sessions that followed were more useful than the product demos, because they named the failure modes the product is designed to prevent.

Before getting into the announcements, something worth naming directly. The conversation about AI in the enterprise is still dominated by the generative AI frame. A large language model receives a question and generates an answer. Enterprises pour budget into that pattern and then are surprised when the answers are wrong, inconsistent, or confidently hallucinated. The reason is sequencing. Analytical AI, which finds patterns in what has already happened, and predictive AI, which forecasts what is likely to happen next, have to come before generative AI can produce anything trustworthy. Generative AI without the first two stages is expensive improvisation. Organizations that double down on generative first are the ones writing off their AI pilots twelve months later and asking uncomfortable questions in board presentations.

Boomi does not frame it that way explicitly. But every product shown across two days is, in practice, an argument for getting the analytical and predictive layers right before asking a large language model to do anything useful with the result. Stan Clark, Vice President and General Manager for AI Market Development at DXC Technology, said it plainly from the stage: if AI is not delivering return on investment, the issue is not the model. It is the foundation.

Boomi Orchestrate Solves the Pilot-to-Production Gap

The new workspace framework inside Boomi Orchestrate is a direct response to the most common enterprise AI complaint: pilots that never reach production. Orchestrate now provides account-level workspaces that isolate and secure digital assets by line of business, so a customer success team and a finance team are not sharing an integration environment without a governance boundary between them. The addition of intent-based Solutions, where a user describes an outcome and Orchestrate proposes the agent architecture, removes the requirement for a developer to translate business intent into a technical specification.

The demo showed a customer support ticket routing agent built through natural language prompting by Ann Maya, Boomi's Global Head of Strategic Projects, with Meta Hub providing business glossary context so the agent understood what "high value customer" meant for that specific organization rather than guessing. That last part matters. An agent reasoning over undefined terms is generating confident fiction. Boomi also previewed Agent SIM, a Labs-stage capability that allows organizations to simulate and validate agent behavior before deployment. In a production environment running twenty-four hours a day, the value of knowing how an agent will behave before it behaves that way is not a nice-to-have.

Venkata Kalikrishna Chekka, Senior Manager of Enterprise Integration Solutions at Suffolk, participated in Boomi's design partner program for Orchestrate and described the value as a more flexible foundation to streamline operations and explore how agent-driven processes can drive efficiency. Design partners who have seen the product under development tend to be the least likely to oversell it.

Knowledge Hub Is the Missing Layer Between Data and Agents

Boomi announced Knowledge Hub as a forthcoming native service on the platform. The positioning is retrieval-augmented generation, the process of grounding a large language model's answers in specific enterprise documents, as a managed capability combining structured and unstructured data sources under Boomi's security and governance framework. The distinction from Meta Hub is worth clarifying. Meta Hub provides semantic definitions, the business glossary that tells an agent what terms mean. Knowledge Hub provides the content those terms point to: documents, incident reports, service level agreement details, policy materials. Agents need both. Definitions without content produce empty reasoning. Content without definitions produces inconsistent reasoning.

The Guru partnership announced today makes this concrete. Guru, which positions itself as an AI Source of Truth for enterprise knowledge, is a launch partner for Boomi Connect. Guru's knowledge agents will draw on Boomi's managed connector layer to reach live operational data at the moment of inquiry, not a cached snapshot from the prior week. Rick Nucci, Guru's Chief Executive Officer and co-founder, put the problem plainly on stage: AI systems fail because of inconsistent, outdated knowledge, not because of insufficient intelligence. Boomi Connect handles the real-time retrieval. Guru handles the verification and governance of what gets retrieved. In regulated industries where a wrong answer has consequences, that separation matters.

Guru noted that many of its customers in financial services and healthcare are eager to adopt AI but face high-consequence accuracy requirements. Knowledge that was accurate last month is not necessarily accurate today. An agent that cannot distinguish between current policy and superseded policy is a liability, not an asset.

Gong in Boomi Agentstudio Turns Conversation Signals Into Enterprise Workflows

The Gong integration announced today puts revenue signals, call transcripts, buyer intent indicators, and competitive intelligence directly into Boomi Agentstudio as live triggers for enterprise workflows. The pattern Boomi demonstrated internally is the clearest version of this: Gong captures product enhancement requests from customer conversations, a Boomi agent routes that feedback to the appropriate engineering backlog, and the workflow closes without a human manually transcribing what a customer said on a call into a ticket. That specific example is not replacing a product manager. It is replacing the twenty minutes of administrative translation work that happens before a product manager can make a decision.

Gong is also registered in the Boomi Model Context Protocol Registry, meaning its revenue signals are discoverable and consumable by any compliant agentic framework on the platform. Eran Aloni, Gong's Executive Vice President of Product Strategy and Ecosystem, described the goal as enabling organizations to take Gong's customer intelligence and drive coordinated outcomes across the enterprise. A pre-built Gong recipe in the Boomi Marketplace handles the Gong-to-Jira product feedback routing workflow for teams that want to start without custom development.

The Automotive Demo Made the Friction Argument Better Than Any Slide

The Mobeus demo was the most direct statement of what Boomi's data activation argument produces for a business user. A prospective car buyer interacts with an AI agent called Penny. The agent checks lease expiry data, pulls inventory from NetSuite, matches the customer's preferred color from Data Hub history, confirms Tesla Supercharger compatibility, and books a test drive in ninety seconds, triggering a calendar invite and a Slack notification to the sales representative. The cost per booking: fifty cents in token spend against an industry average of one hundred fifty dollars per booked test drive.

What made the demo more interesting than the outcome was what Richie from Mobeus described about how the agent selected which features to sell. The agent was evaluating eighteen possible selling points simultaneously. It surfaced vehicle speed because the behavioral data said that would close this customer. Nobody programmed that selection. The agent reasoned over connected data and made a call. That is analytical AI feeding the generative response, which is precisely the sequencing argument.

The entire connectivity was wired up in fifteen minutes over a screen share. The rest of the ninety seconds was story.

HNL Lab Medicine: Six Months, Twenty Platforms, Three Million Patients

The most operationally grounded customer story across two days came from Eric Brown, Vice President of Information Services and Technology at HNL Lab Medicine, one of the largest independent laboratory networks in the United States. HNL serves more than three million patients and approximately twenty-five thousand providers annually. Before Boomi, the environment was a collection of siloed platforms: an information system, Epic, multiple electronic medical record systems, and others that were not exchanging data. Integration work was manual scripting, non-standard, and not scaling.

Six years into the Boomi relationship, HNL integrated more than twenty platforms and two hundred workflows within the first six months. The impact Brown described was not operational efficiency in the abstract. It was patient care. In a laboratory environment, the time between a critical result and the provider who needs to act on it is a clinical variable. Manual handoffs introduce delay. Delayed results introduce risk. Boomi eliminated the manual steps in that chain.

HNL is now planning to triple in size over the next three years, expanding partnerships with larger health systems across the East Coast. The integration foundation they built is the prerequisite for that growth. Brown's advice to other healthcare and enterprise organizations considering a transformation at similar scale: look at the problem first, and do not worry about the technology. The technology fills in the gaps once the problems are clearly understood.

That is a practitioner framing that most AI vendors do not want to hear. It puts the buyer's problem-definition work ahead of any platform selection conversation.

DXC Said What Most Vendors Will Not

The most analytically useful session of Day 2 was the conversation between Ed Macosky and Stan Clark, Vice President and General Manager for AI Market Development at DXC Technology. DXC runs core systems for banks, manufacturers, and governments across the world, which means Clark sees what happens when AI projects meet the actual enterprise environment rather than a controlled demo.

Clark's estimate: by some measures, eighty percent of AI projects are failing or never reaching production. His diagnosis was precise. Enterprises were not built for agents. The data foundation, the integration layers, and the documented processes that agents need to operate were not designed with agents in mind. The people inside those enterprises became the connective tissue that compensated for what the systems could not do. When an agent is deployed into that same environment, it does exactly what it is told, repeatedly, without judgment. That is not a problem. Clark called it the first honest X-ray of your business operations. The agent reveals what is actually happening, including the workarounds and the gaps that the humans were quietly managing.

His three prerequisites before any enterprise AI deployment are worth keeping: data must be reachable, connected, and trusted; the integration layer must be solid because agents live on application programming interfaces and a fragile API stops an agent cold; and processes must be documented clearly enough that an agent can execute them without asking clarifying questions, because agents do not ask questions. They decide and they act.

Clark also asked four governance questions that every organization should answer before deploying any agent into production: When does it decide autonomously? Who is accountable when it is wrong? How do you control it mid-flight? And who can stop it? Most organizations deploying agents today have not answered all four. The ones that have not are the ones building the eighty percent statistic.

His framing on the destination is the part that most AI vendor conversations skip. Cost reduction and efficiency gains are the entry point. They are where the business case starts and where the board presentations focus. But they are not the definition of success. When loans are approved in minutes, when claims are validated instantly, when customers are helped while they are still asking the question, the conversation shifts from what can we save to what can we create that we could not create before. That includes the new markets that were previously uneconomical, the personalized offers that required more manual effort than they returned, and the products that could not be built because the people who would have built them were processing transactions instead.

Boomi Companion Changes Who Can Build Integrations

Boomi Companion, announced by Matt McClure, is a set of agent skills and plugins that converts any compatible agentic engineering environment into a Boomi expert. The demo used Claude Code. An email containing Stripe integration requirements and documentation was dropped into Claude Code with the Companion skills installed. Eight minutes later, Companion had built a complete Boomi process with field mappings, error handling, logging steps, and a reused existing connection, then ran its own tests before returning the result. McClure was explicit that Companion encodes actual Boomi engineering best practices, not a skeleton process requiring expert cleanup. The output was what a seasoned Boomi professional services engineer would build, including the monitoring and introspection steps that make a process testable in production.

The barrier to Boomi integration work has historically been the availability of certified Boomi developers. Companion shifts that constraint without eliminating the underlying expertise requirement. It compresses the time required to apply expertise that still has to exist somewhere in the organization.

Companion works with Claude Code, OpenAI Codex, and Microsoft Copilot. The skills are also available in non-developer environments for documentation generation, including multilingual output. McClure demonstrated English and Japanese documentation generated from the same process. For enterprises operating in multiple regions, that is not a minor feature.

Anomaly Detection, Agent Isolation, and the Governance Gap

The Agent Control Tower added proactive anomaly detection using thirty days of behavioral data to establish baselines for each agent. The demo showed a flagged critical issue: suspicious multi-region access with a high failure rate. The response was agent isolation by region through agents in runtime, which restricts a specific agent's execution to designated environments without taking the whole workflow offline. For enterprises in regulated industries, this is the difference between a governance incident and a governance catastrophe.

Boomi also announced Agentstudio Multi-region Instances, which allows organizations to leave agent metadata and runtime execution in specified regions. The distributed agent runtime capability deploys agents on-premises with locally hosted language models, keeping sensitive data behind the firewall and reducing cloud inference costs. These are not edge features. They are the requirements that healthcare, financial services, and public sector buyers put at the top of any evaluation criteria.

Boomi Embed Kit was the quieter announcement. It allows a third-party application to surface Boomi-governed data access through a lightweight embedded interface, so a user in an expense management application can ask a natural language question and receive an answer drawn from internal policy documents, without leaving the application and without the data leaving Boomi's governance layer. The on-stage scenario involved an employee trying to find their expense policy limit before taking clients to dinner. The agent could not answer because it had no access to internal documents. Embed Kit is the fix for that specific failure mode, which plays out across thousands of enterprise applications every day.

The Linux Foundation Membership Is a Standards Signal

Boomi announced membership in the Linux Foundation, specifically joining the Agentic AI Foundation, which governs specifications including Model Context Protocol and agent interoperability standards. A vendor that participates in the standards body for the protocols its platform depends on has a different accountability relationship to those standards than one that implements them selectively and without community review. This matters for enterprise buyers evaluating platform lock-in risk over a three-to-five year horizon.

Boomi also launched Agentic Academy, a public resource for best practices around combining deterministic and agentic processing. Blending rule-based workflows with AI agents is where most enterprise deployments break down: organizations treat agents as a replacement for deterministic logic rather than a complement to it. Clark and Macosky both made this point from different directions. Deterministic processing handles the things you know. Agentic processing handles the things you do not know until they happen. Mixing them without a clear boundary is how you get agents doing things nobody expected them to do.