# Dataiku Agent Management: Track Your AI Agents Now > Source: > Published: 2026-09-25 20:12:49+00:00 Most enterprises are confident they know which AI agents they’re running. Most are wrong. Research published this week by Guild.ai found that 96.4% of IT decision-makers believe they maintain a complete agent inventory — yet only 42.7% have centralized monitoring tools in place. Dataiku shipped Agent Management on September 24 to address exactly that gap: a standalone, platform-agnostic product that discovers, tracks, and risk-ranks every AI agent running across your stack, regardless of which vendor built it. General availability is October 2026. ## The Problem: Agents Multiply Faster Than Anyone Can Count The numbers are stark. IBM projects more than [1,600 AI agents per large enterprise by year-end 2026](https://www.ibm.com/think/topics/ai-agent-sprawl). Gravitee’s State of AI Agent Security 2026 counts over three million agents operating inside corporations globally, with fewer than half actively monitored. Meanwhile, 66.7% of organizations with agents have already experienced an agent-related operational consequence — a failed customer interaction, an unauthorized data access, a cascading workflow failure — in the past 12 months. Dataiku CEO Florian Douetteau put it plainly: “Ask a bank how many servers it runs, and you get an answer to the decimal. Ask how many AI agents it’s running, and you get a shrug or a guess.” That shrug is now a liability. Sixty percent of organizations suspect employees are deploying agents without IT approval. Nearly half have decommissioned more than 20 agents this year — quietly, without documentation, without understanding why those agents failed. ## What Dataiku Agent Management Does Agent Management connects to your existing agent infrastructure via native APIs and log streams. You do not need to rewrite or redeploy your agents to gain visibility. Once connected, it automatically inventories each agent’s structure — which tools it uses, which models it calls, which data sources it can reach. It then tracks performance against both technical metrics and business KPIs, and classifies each agent by risk tier: agents handling customers, sensitive data, or live transactions get flagged for mandatory certification and scheduled testing. The product also supports natural language queries across your entire portfolio. Platform coverage is the core pitch. Agent Management connects natively to AWS Bedrock, Databricks Agents, Google Vertex AI, Microsoft Copilot Studio, Azure Foundry, Salesforce Agentforce, and Snowflake Cortex. For agents built on custom frameworks, OpenTelemetry instrumentation fills the gap. The depth of integration scales with what each platform exposes, and Dataiku says deeper integrations are actively in development. ## Open Source vs Commercial: Two Approaches to the Same Problem Dataiku is not the only governance product shipping this month. [WSO2 launched Agent Manager on September 15](https://www.infoq.com/news/2026/09/ws02-agent-manager/) — fully open source under Apache 2.0 — targeting the governance and policy enforcement layer rather than portfolio visibility. WSO2 controls what agents can do: verifiable agent identity, role-based access, PII masking, rate limiting, sandboxed Kubernetes runtime. Dataiku shows you what agents are doing: business metrics, risk classification, audit trails. These sit at different layers of the same stack, and the choice between them is not either/or. | | Dataiku Agent Management | WSO2 Agent Manager | |---|---|---| | License | Commercial | Apache 2.0 (open source) | | Focus | Portfolio visibility & business metrics | Runtime governance & policy enforcement | | Deployment | SaaS managed | Self-hosted or managed SaaS | | Available | October 2026 | Now (GA) | Five agent governance products shipped in 12 days in mid-September 2026. [Guild.ai’s research](https://www.globenewswire.com/news-release/2026/09/22/3366611/0/en/the-ai-agent-management-gap-is-growing.html) captured the underlying demand: 83.9% of organizations view real-time agent visibility as extremely or very valuable, and 81% consider centralized management hubs a critical need. NiCE acquired Cognigy for $955 million in the same window. The enterprise AI control layer is forming fast, and it is fragmenting by function — routing, runtime governance, and portfolio visibility are becoming distinct product categories. ## What to Do Before October GA You do not need to wait for GA to start. Three practical steps now: 1. **Run an informal audit.** List every AI agent your team has deployed across every platform. Include agents your organization did not build — agents spun up by other teams on Copilot Studio or Agentforce count. The number will likely surprise you. 2. **Classify by risk tier.** Any agent with access to customer data, production databases, or financial systems is high risk. Tag it now. These are the agents that need certification frameworks in place before October GA, not after. 3. **Instrument with OpenTelemetry.** If your custom agents are not already emitting OpenTelemetry traces, start now. Both Dataiku and WSO2 use it for custom environment connectivity. It is the emerging standard for agent observability — the same role Prometheus took for infrastructure metrics a decade ago. We covered [WSO2 Agent Manager](https://byteiota.com/wso2-agent-manager-open-source-control-plane/) when it shipped ten days ago, and [Bedrock AgentCore with GitHub Actions](https://byteiota.com/bedrock-agentcore-github-actions-agent-ci-cd/) the day before that. The pattern is clear: agent governance is the new DevOps frontier, and the tooling is arriving fast. The agents are already in your enterprise. The question is whether you find them before they find you.