Sitting at Gate C26 waiting for the flight home from Chicago, the number that keeps coming back is not 90,000 agents or 30,000 customers. It is seven percent. Only seven percent of enterprise data is in motion today, according to Steve Lucas in the Day 1 keynote. That is the constraint the entire event was organized around, even when nobody said it directly. Every product announcement, every customer story, every governance argument traced back to that gap between the data enterprises own and the data they can actually use.
I covered the infrastructure announcements on Day 1 and the product sessions on Day 2. This third piece is the closing argument: what the two days added up to, and a framework I came away with for thinking about what enterprises actually need before any agent conversation makes sense.
The Six Things Every Enterprise Needs Before the Agent
Boomi frames its platform journey in three stages: Connected Enterprise, Automated Enterprise, Agentic Enterprise, all sitting on an Active Data foundation. That is the right architecture. What it understates is the sequencing work inside each stage. Stepping back across two days of sessions, six things kept surfacing as the prerequisites that actually have to be in place before the agentic layer delivers anything durable. This is my reading of the pattern, not a Boomi-authored framework.
Every system the agent will need has to be reachable through governed connectors before a single workflow is built. Boomi ships 1,500+ connectors. The count matters less than the governance layer on top of each one.
Moving data from rest to motion. Lucas put the current baseline at seven percent. Enterprises that cannot deliver governed data to human intelligence in real time are not ready for agents operating at machine speed.
Analytical AI, finding patterns in what has already happened, has to precede generative AI. DXC's Stan Clark made this point across two different sessions without ever using the phrase analytical AI. The behavioral data that let the Mobeus agent select which of eighteen selling points to surface was the output of analysis, not generation.
Adding meaning to data through the Context Layer. Data Hub provides clean governed data at the source. Meta Hub provides shared definitions that agents trust. Knowledge Hub, coming next, adds the document and policy layer that grounds agent reasoning in enterprise-specific content rather than general inference.
Shifting from reactive to proactive. Agent Control Tower's new anomaly detection uses 30 days of behavioral baselines to flag deviations before they become incidents. That is not a monitoring feature. It is what makes it possible to trust an agent operating without constant human supervision.
Only here does the agentic layer do anything useful. Boomi Orchestrate, Agentstudio, and Boomi Connect are the execution layer. They are real and capable. The question is whether the five steps above are solid enough to support what runs on top of them.
Lexitas Made the Foundation Argument Concrete
The customer case that stayed with me was not the largest or most technically complex. Lexitas, a legal services company, had over 2,500 payment lines processed manually every day, with data scattered across ERP systems, subsystems, and shared mailboxes. One to two dozen specialists were required to reconcile each payment. That is not an AI problem. That is a data activation problem that an agent can only address after the foundation work is done.
Agentstudio built the payment processing agent end-to-end. Data Hub unified all systems so the agent could reason across them. Agent Control Tower provided full audit trail and governance. John Baker, CIO and CISO at Lexitas, described the operating principle on stage: pick something bite-sized, do it quickly, then build from there.
What the Lexitas numbers represent is not a generative AI story. It is a data activation story that ends with an agent doing something useful because the first four steps in the framework above were already done.
"AI is too strategic to wait on. Pick something bite-sized, go do it, and do it quickly — then build from there."
John Baker, CIO & CISO, Lexitas — Boomi World 2026The Honest Number Nobody Puts in the Press Release
Stan Clark from DXC Technology estimated that 80 percent of AI projects are failing or never reaching production. He made that statement on Boomi's own stage, in front of Boomi's own customers. The reason Boomi invited that assessment is because their product argument depends on it being true. If the foundation work is not the bottleneck, Boomi is just a connector business. If it is, Boomi is the layer that makes the 80 percent avoidable.
Clark's four governance questions, asked before any enterprise deploys an agent, are worth repeating. When does the agent decide autonomously? Who is accountable when it is wrong? How do you control it mid-flight? Who can stop it? Most organizations deploying agents today cannot answer all four. That is not a technology problem. It is a readiness problem that no platform resolves on its own.
The Boomi Companion Detail Worth Watching
The Day 2 demo of Boomi Companion building a complete Stripe integration in eight minutes, including field mappings, error handling, and logging, using Claude Code as the development environment, was the quieter announcement with longer-term implications. The constraint on Boomi deployments has historically been certified developer availability. Companion shifts that constraint without eliminating it. It compresses the time required to apply expertise that still has to exist somewhere.
For enterprises with Boomi footprints and a thin bench of integration developers, that is a meaningful change to what they can build without growing headcount. For enterprises evaluating Boomi for the first time, it changes the onboarding economics.
Available now: Agentstudio, Agent Control Tower, Meta Hub, Boomi Companion, 1,500+ governed connectors, European Platform Instance. Launching: Boomi Connect, MCP Registry, Boomi Orchestrate, Agentstudio APIs, EmbedKit CDN. Coming next: Knowledge Hub, Agents in Runtime, AI Gateway, Agent Context Engineering, BYOM Agentstudio.
