IBM and Microsoft announced a joint platform today, and the framing is the product. Imagine Enterprise, they say, is for organisations that want to "build, run and govern" AI agents — and the word govern is carrying as much weight as build.
The mechanics are a division of labour. IBM Consulting brings more than 6,000 pre-built AI agents and assets built on its Consulting Advantage delivery platform, plus forward-deployed engineers to install them. Microsoft brings the substrate: Foundry, GitHub Copilot and Copilot Studio to coordinate agents, and Fabric, Microsoft 365, Entra, Purview, Defender and Azure for data, identity, governance, security and cloud. IBM adds Process Studio, Context Studio and Advantage Core for designing, deploying and governing tailored agents.
The 6,000 figure needs context. It is IBM Consulting's catalogue of agents and assets accumulated across client engagements — not a shrink-wrapped product line — and no independent benchmark accompanies it. What is genuinely new is the packaging: one offering in which a regulated enterprise does not have to assemble agents, identity, data plumbing and compliance tooling itself, backed by engineers from both companies for qualifying engagements. The industry frameworks available at launch are financial services, insurance and retail, with manufacturing, energy, public sector and mobility named as future candidates. Those three are the sectors where an ungoverned agent is not a productivity story but an audit finding.
The customer named is SEI, the financial technology and asset-management services firm with roughly $2 trillion in assets under management, administration or advice as of 30 September. Its chief AI strategist, Sneha Shah, describes work exploring emerging capabilities "with intention" — the language of a pilot, not a rollout. That is normal for day one, and it is also the only evidence offered.
The strategic logic is legible on both sides. IBM Consulting is betting that the durable position in the agent economy is systems integrator rather than model owner: sell the deployment, the governance and the industry vocabulary, stay model-agnostic, and let somebody else eat the capex of training. Microsoft gets distribution into IBM's regulated client base without building a consulting army of its own. Mohamad Ali, who heads IBM Consulting, frames the goal as turning "AI from a productivity aid into a real competitive advantage"; Microsoft's Rodrigo Kede Lima calls it turning "AI ambition into business impact".
Neither framing is a number. The announcement contains no pricing, no agent-level performance data, no count of customers in production, and no statement about which models the agents actually run on. That last absence is the most interesting. A platform whose entire selling proposition is governance is silent about the component regulated buyers most want to pin down — and "model-agnostic" is a promise that sounds like flexibility until a supervisor asks which model produced the decision.
Read against the rest of this week, the move fits a pattern. Model vendors are racing downmarket with cheap, agent-optimised models; the integrators are racing to own the layer where those models meet a compliance department. Imagine Enterprise is IBM's claim on that layer with Microsoft's cloud underneath it — a hedge against the possibility that models commoditise and the money moves to whoever can prove the agent behaved.
The governance claim deserves precise interrogation, because there are two very different things behind the word. One is enforcement: policy that blocks an agent from taking an action. The other is observability: logs and traces showing what it did after the fact. Purview and Defender are strong at the second; the first is where regulated industries actually get hurt, and it is the part of the stack that agent platforms most often leave to the customer.
What to watch: whether the initial verticals produce named production deployments rather than explorations, whether a price appears, and whether "governed" turns out to mean enforced policy or merely observed behaviour. Observability is not control, and the difference is the whole argument.

