Box, Domo, and SAP Are All Making the Same Bet: Enterprise AI Needs a Native Source of Context
Three enterprise platforms repositioned themselves as agent infrastructure in the same week. The common thread: agents need context from the systems that already own it.
What Happened
Within the span of a few days, three enterprise platforms moved in the same direction.
On April 2, Box made Box Agent generally available for Enterprise Plus and Enterprise Advanced customers — an AI capability that takes natural-language instructions, searches enterprise files, extracts data, and generates structured outputs (Word docs, PDFs, spreadsheets, slides) without leaving Box's permissions framework (Techzine).
Domo announced its AI Agent Builder, AI Toolkits, a centralized AI Library, and a Domo MCP Server — tools designed to let external AI assistants like Claude, Gemini, and ChatGPT query governed business data, trigger workflows, and run operational processes through a single connected surface (Demand Gen Report).
And SAP acquired Reltio, a master data management platform, to bring entity resolution and clean "golden records" into SAP's Business Data Cloud — and by extension, into Joule and Joule Agents (Techzine).
Three different vendors. Three different data types. One clear pattern.
Why This Is Bigger Than Three Product Launches
The standard read here is "incumbent platforms adding AI." That's the wrong read.
What's actually happening is an architectural shift. For the past two years, enterprise AI has largely been deployed as a separate layer — a chat interface bolted on top of existing systems. You'd get a general-purpose assistant, then separately connect it to your data. The integration was the product.
That model is changing. Box, Domo, and SAP aren't building AI on top of their platforms. They're making the platform itself the place where agents get their context, their permissions, and their ability to act. The architecture flips: instead of AI reaching into your systems, the systems reach back.
Box CEO Aaron Levie put it plainly: "AI can only reach its full potential if it understands the unique context of an organization." That context, he said, lives in contracts, research materials, and financial documents — exactly what Box already holds (Techzine).
SAP's framing was almost identical. "AI cannot reach its full potential when data is fragmented across business units, platforms, and domains without connection or context," said Muhammad Alam, SAP Product & Engineering, explaining the Reltio acquisition (Techzine).
When two of the largest enterprise software vendors say essentially the same thing in the same week, it's worth sitting with.
Three Platforms, One Playbook
Box: Unstructured Content as an Agent Work Surface
Box Agent applies reasoning models from OpenAI, Anthropic, and Google to unstructured enterprise documents. It handles multi-step tasks in a single flow — locate files, extract data, generate outputs — all inside Box's existing permission controls.
The practical scope is wide. Legal teams can compare contracts against standard playbooks. Sales engineers can automate RFP responses by pointing the agent at existing proposals. Marketing can draft brand-consistent assets grounded in existing style guidelines. None of this requires a separate integration; it runs where the content already lives.
The critical design choice: the agent is permissions-aware by construction. It only accesses files the user is authorized to see, and customer data is not used to train third-party models. Enterprise AI adoption has stalled repeatedly on exactly these concerns. Box is betting that native context plus native permissions is how you clear that hurdle.
Domo: Governed Data as an Agent Interface
Domo's framing is "AI as a system of action." The AI Agent Builder lets teams create conversational agents and agentic workflows scoped to specific use cases. AI Toolkits define what each agent can do. The AI Library provides a centralized hub for managing these agents at scale — available to customers this summer.
The most significant piece is the Domo MCP Server. By implementing the Model Context Protocol, Domo opens its governed business data and workflows to external AI platforms without requiring those platforms to own the data. External assistants can query data, trigger workflows, create dashboards, and run operational processes against Domo's governed context.
This is a deliberate design choice. Rather than locking customers into a single AI vendor, Domo is positioning itself as the context layer that any assistant can connect to. The bet is that whoever controls trusted business context wins — regardless of which model the customer ends up running.
SAP + Reltio: Clean Records as Agent Prerequisites
SAP's move is more structural than either of the others. Reltio's master data management platform uses AI-based entity resolution to identify and merge related data records across different formats, systems, and domains — producing a single reliable "golden record" for each business entity.
The acquisition pulls this capability into SAP Business Data Cloud, making clean, connected business records a foundation for both the Joule assistant and the Joule Agents ecosystem. Reltio also supports MCP, enabling real-time multi-agent workflows across systems. Pre-built "velocity packs" for life sciences, healthcare, and financial services are included.
SAP's stated rationale is the clearest summary of what all three vendors are doing: AI fails when business data is fragmented. The answer isn't a better model. It's better data infrastructure — consistent, attributed, maintained.
What Enterprises Should Notice
The shift from "AI on top" to "AI embedded in the system of context" carries practical consequences that most AI adoption conversations skip.
Actions, not just answers, are the stakes now. When an agent takes an action — generating a contract, triggering a procurement workflow, flagging a supplier — the quality of the context it read becomes an execution risk. A bad answer costs attention. A bad action costs money, relationships, or compliance standing.
Context ownership is becoming strategic. The platform an enterprise trusts to hold its canonical documents or business records increasingly controls what the agent knows. That makes the knowledge layer a competitive moat, not an infrastructure afterthought.
Permissions and provenance are table stakes in production. Box's permissions-aware design and SAP's entity-resolution layer both exist for the same reason: enterprise buyers won't deploy agents that might read files they shouldn't, or act on data with unclear provenance. This is no longer a differentiator — it's a prerequisite.
The Problem These Platforms Haven't Fully Solved
Something deserves to be said plainly: exposing documents or data to an agent is not the same thing as maintaining a trustworthy knowledge layer.
Box Agent can find and synthesize what's in Box. But if Box holds both an outdated policy and its replacement — which happens constantly in real enterprises — the agent doesn't know which to trust. Reltio can produce a golden record from fragmented business data, but a golden record built from stale inputs is just a single point of failure dressed up as a solution.
The harder problem in enterprise AI isn't context access — it's context quality. Documents decay. Policies get updated in one system and forgotten in three others. Contradictory information accumulates quietly until an agent acts on it.
This is the gap that a governed knowledge layer is built to close. Source-attributed retrieval, contradiction detection across documents, automated knowledge maintenance, scanned-PDF handling — these capabilities exist to answer not just "can the agent read this" but "should the agent trust this." The flurry of platform announcements this week is the market confirming that the question matters. Mojar AI is built around the answer.
What to Watch
More platforms will follow. Any enterprise system that owns a meaningful slice of business content, records, or structured data has an incentive to position itself as an agent foundation. MCP adoption will accelerate this — cutting the integration cost of connecting agents to new context sources.
The next competitive divide won't be between platforms that support agents and those that don't. It will be between platforms that can keep their context trustworthy enough to act on, and those that can't.