Oracle tops the agentic AI analyst chart as CrewAI ships a runtime control plane and shadow AI goes mainstream

The analyst verdict

Oracle opened the week with the credential enterprise buyers notice first: a Leader placement in the IDC MarketScape: Worldwide Agentic AI Platforms 2026 Vendor Assessment. The positioning is deliberate — Oracle's OCI Enterprise AI is framed as a unified platform that folds model access, agent development, orchestration, and governance into one managed service, with support for both Model Context Protocol and Agent2Agent so agents can connect to tools and to each other. The message to procurement teams is that governance arrives bundled, not bolted on. m:…8468

Oracle's cited customer proof point is ARENCO Group, which the company says used OCI Enterprise AI to automate document checks — one concrete workload, not a transformation narrative. And the company stresses model flexibility: enterprises keep choice over models while Oracle holds the security and operational controls. It is a company telling its own story, so read the Leader placement as a vendor's best case, not a buyer's conclusion. m:…8468

The runtime answer

CrewAI took the opposite tack from the analyst chart: build the control plane inside the runtime. Its Agent Control Plane, launched on the CrewAI AMP platform, governs agents built in its low-code Crew Studio and agents written in Python with CrewAI open source — one enforcement surface across both. The idea is to put policies such as PII masking, cost limits, and tool restrictions directly into the execution loop, since external HTTP proxies cannot inspect an agent's internal state transitions. m:…6509

The honest part of the announcement is the scope note: CrewAI says the Agent Control Plane does not replace identity layers or systems of record, and it ships with 13 out-of-the-box policies for platform administrators. That is the right boundary to state — a runtime policy engine is not an IAM system — but it also marks what buyers still have to assemble themselves. Governance inside the loop is a real product advance; it is not the whole compliance answer. m:…6509

The problem both are chasing

Dataiku's argument this week explains why both announcements exist. Shadow AI — any AI tool, model, or agent used at work without IT's knowledge or sign-off — is already inside the building. Dataiku cites figures that make the point sharply: 71% of UK employees reported using unapproved consumer AI tools at work in October 2025, and one in five organizations reported a security breach tied to shadow AI in 2025. The numbers are cited by a vendor selling the answer, so treat them as directional rather than definitive. m:…7162

The diagnosis is the interesting part: analysts bypass official platforms because approved tools are too rigid or take too long to access. Dataiku's prescription follows from that — don't add stricter controls, provide faster governed platforms, with tools like its Cobuild letting teams build, deploy, and audit agents in a visible framework. Whether speed actually beats shadow AI is unproven in the evidence. But the framing reframes the whole governance debate: the ungoverned agent is not a policy failure, it is a product failure — the approved path was too slow. m:…7162

Why it matters

Put the three together and enterprise AI's center of gravity has shifted: governance is no longer the slide at the end of the deck, it is the product. Oracle sells it bundled in the platform, CrewAI sells it inside the execution loop, Dataiku sells it as the reason the fast path must be the governed path. The question for buyers is no longer whether agents are governed — every vendor now says yes — but where the enforcement lives and what it still doesn't cover. Ask which layer each vendor's governance replaces, and which it quietly assumes someone else owns. m:…8468 m:…6509 m:…7162