On 10 November 2026 IBM and Microsoft announced Imagine Enterprise, an agentic platform for building, running and governing AI agents for industry-specific processes. It combines more than 6,000 pre-built AI agents and assets from IBM Consulting with Microsoft’s AI, data, cloud and security stack, and is aimed at regulated industries, according to the joint announcement.
The first industry frameworks cover financial services, insurance and retail. Manufacturing, energy, public sector and mobility are named as future targets.
/ 01What was announced
On the IBM side, the platform draws on IBM Consulting Advantage, the firm’s AI-powered delivery platform, including tools called Process Studio, Context Studio and Advantage Core for designing, deploying and governing agents. On the Microsoft side it uses Microsoft Foundry, GitHub Copilot, Copilot Studio, Fabric, Entra, Purview, Defender and Azure. The companies describe built-in controls, observability, AI-compliance support, and enterprise security and identity.
There are three ways to use it: IBM Consulting deploys and runs it for the client, the client co-develops with IBM Consulting, or the client manages it themselves. Forward-deployed engineers from both companies support qualifying engagements. The named early customer is SEI, which is exploring agentic technologies with IBM Consulting with Microsoft’s support.
Mohamad Ali, head of IBM Consulting, said: “Clients are no longer looking for AI that simply helps people do the same work a little faster.” Rodrigo Kede Lima, president of Microsoft’s Frontier Company, said: “Enterprise AI will be won by organizations that can turn their data, industry expertise, and AI into action.”
AI Weekly observed that the release does not disclose pricing or other pilot customers, and does not explain how individual agent actions are logged for auditors. It also characterised the launch as largely a services offering, since access runs through IBM Consulting engagements.
/ 02Why it matters
Two of the largest enterprise technology firms are packaging agents by industry and selling governance as part of the product. That reflects what buyers in regulated sectors have been saying: a general-purpose assistant is not enough, and an agent that touches claims, payments or customer records has to be explainable to a compliance team.
The 6,000 pre-built agents are the headline, but the more telling element may be the tool IBM calls Context Studio. Pre-built agents encode generic industry processes. Every insurer, bank and retailer runs those processes differently, with its own product definitions, exceptions and rules. An agent that does not know those differences will produce answers that look right and are subtly wrong. The value of the platform will depend on how well each client’s own context is captured, not on how many agents ship in the box.
/ 03Governance means being able to replay a decision
AI Weekly’s point about logging is the question regulated buyers should press. Observability dashboards show that an agent ran, how long it took and what it cost. Auditors want something else: for a specific decision, what information the agent used, which rule or policy it applied, and whether the same inputs would produce the same result.
That requires the business context to be explicit and versioned. When products, policies, customers and obligations live in a governed knowledge graph, an agent’s decision can be recorded as a path through specific, dated facts. The path can be replayed. A reviewer can see that a claim was declined because policy clause X, as in force on date Y, applied to coverage Z, rather than because a model found the denial plausible. Deterministic reasoning over that graph makes the answer reproducible, which is what a regulator means by auditable.
/ 04What to ask before buying any agent platform
- How is our context captured? Who models our products, policies and terms, and in what form? Can we export it?
- What does an audit record contain? Ask for a sample decision trace, not a dashboard.
- Is the reasoning reproducible? Same inputs, same answer, same evidence.
- Where does it run? For sensitive data, whether the platform can run inside your own perimeter.
- What happens when the model changes? Whether agents keep behaving the same way when the underlying model is updated.
Whatever the scale of the vendor, the questions above are the ones that separate an agent demo from an agent a regulator will accept.
Six thousand pre-built agents is a catalogue, not a control framework. For an insurer or a bank, what matters is whether any one of those agents can reproduce a specific decision for an examiner months after it was made.
That requires the client’s own context, its policy wordings, product terms, exceptions and obligations, to sit in a versioned graph the agents read from. A claim decision then records a path: claim C, coverage Z, clause X as in force on date Y, exclusion applied, outcome declined. Re-run it with the same inputs and deterministic reasoning returns the same answer with the same evidence; amend the clause and the graph shows which past decisions relied on it. Generic agents become specific by binding to that graph, which is how our Agent Catalog is meant to be used: prebuilt clinical, business and intelligence agents adapted to an organisation’s own knowledge and deployed behind its firewall, with the context model exportable as OWL or RDF.
How we’d model it
Claim decision—applied→Policy clausePolicy clause—inForceDuring→Policy versionClaim—coveredBy→CoverageAgent action—authorizedBy→Underwriting rule
- IBM and Microsoft Announce a New AI Agent Platform Designed for Industry-Specific Transformation — IBM Newsroom newsroom.ibm.com ↗
- IBM, Microsoft launch Imagine Enterprise with 6,000 agents — AI Weekly aiweekly.co ↗
See how Aktiver turns raw data into knowledge graphs your AI can reason over — deterministic, traceable, behind your firewall.