Gartner ranks AI agent management platforms: what NZ firms need

What Gartner published

Gartner has released its first Magic Quadrant for AI agent management platforms, and Microsoft dates the report to 27 October 2026. Vendors spent the first days of November announcing where they landed. According to CX Today's 2 November coverage, Gartner rated Microsoft, IBM and ServiceNow as Leaders. Amazon Web Services, Salesforce, OneReach.ai and Gravitee are Challengers, Boomi and Automation Anywhere are Visionaries, and UnifyApps, Dify, AvePoint and Lyzr are Niche Players.

Gartner defines these products as "centralized tools for managing AI agents regardless of where they run." CX Today lists the capabilities Gartner looked for: agent discovery, lifecycle management, guardrail enforcement, evaluation, identity and access management, integration controls and cost tracking. The same article reports a Gartner prediction that 60% of AI agents will be managed through an agent management platform by 2029.

Gartner's analysis sits behind a paywall, so most of what's public comes from vendor announcements, and those come with their own sales pitch. Microsoft's 3 November post says more than 50 million agents have been registered in Agent 365. The product is generally available with Microsoft 365 E7 or as an add-on to E5. Agent 365 already registers agents built on Agentforce, Amazon Bedrock, Anthropic, Databricks, Google Gemini and Google Vertex, and Microsoft says OpenAI, ServiceNow and Snowflake support is coming. Microsoft is also adding 35 local agent types to its discovery feature, which brings the total to 60. A new Action Center lets administrators clean up inactive agents, reassign agents with no owner and block agents that look risky. IBM's 2 November announcement emphasises deployment across SaaS, sovereign and air-gapped environments, and says watsonx Orchestrate can manage agents built on other vendors' technology.

Why a management category has appeared now

Most organisations didn't plan to run dozens of agents. The agents built up as features inside software the organisation already licenses, and in many cases no single person approved them. Salesforce's 2026 Connectivity Benchmark, as reported by IT Brief Australia in February, found Australian organisations run an average of 11 AI agents. Half of those agents work in silos, and only 54% of the organisations have a centralised governance framework with formal oversight of agentic capabilities. The same research found 96% of IT leaders worry agents will add more complexity than value.

Gartner's report also coincides with growing interest in finding AI use that nobody approved. On 4 November, Gigamon released network tools that classify generative AI and large language model traffic so security teams can separate sanctioned from unsanctioned use. Vendors have started selling discovery as a product because organisations can't produce a complete agent inventory on their own.

What it means for NZ businesses

Few New Zealand mid-sized firms will buy a dedicated agent management platform this year, and the Magic Quadrant is weighted towards large enterprises. Gartner's capability list is still useful to smaller firms as a checklist, because it describes the questions a regulator, insurer or board will eventually ask about any agent with access to company systems.

The Privacy Act already sets some of those expectations. The Office of the Privacy Commissioner says the Information Privacy Principles apply to anyone using AI tools in New Zealand. The Office expects a privacy impact assessment before an AI tool is used, regular updates to that assessment, and secure handling of prompts and training data under IPP5. A business can't assess an agent its staff switched on without telling anyone, so the inventory work Gartner describes is also the first step towards meeting the Commissioner's expectations.

Licensing deserves a careful look. Microsoft's post names E7 and the E5 add-on as the routes to Agent 365 and doesn't mention other licence tiers. Firms on Business Premium or E3 should confirm what Agent 365 would cost them before assuming the product is within reach. Firms whose agents run mostly in Salesforce, ServiceNow or a line-of-business application should check how much of their estate a Microsoft-centred registry would actually see today, given that several integrations are still listed as planned.

Vendor neutrality is the bigger question. The three Leaders also sell the agents that need managing, and each has an obvious interest in becoming the control point for the rest of a customer's estate. CX Today's assessment is that buyers should judge vendors on whether the platform can govern every agent the organisation already runs. For a New Zealand firm with a few Copilot agents, a CRM assistant and some automation scripts, the answer to that question will vary a lot between products.

Practical next steps

• Build an agent register now, even if it's a spreadsheet. Record each agent's name, business owner, purpose, the systems and data it can reach, and who approved it.
• Check which agent features are already enabled in the platforms you license, including Copilot Studio, CRM assistants, service desk tools and finance systems. Turn off or document anything nobody can explain.
• Give every agent a named owner and its own identity with the minimum access it needs. Retire agents whose owner has left the business.
• Run or update a privacy impact assessment for any agent that touches personal information, following the Privacy Commissioner's AI guidance.
• Track cost and a measurable outcome for each agent so you can decide which ones are worth keeping.
• Before you commit to a management platform, test it against at least one agent built on a different vendor's stack, and ask for written confirmation of which integrations are live and which are still on the roadmap.

If you're starting to deploy agents, DataForge's AI agent services cover design and integration, and our AI consultancy work covers governance and privacy assessment for NZ and Australian organisations.