Dataiku launches Agent Management to inventory AI agents across platforms

24 Sep 2026 (ET): Dataiku unveiled Agent Management at Dataiku Succeed, a standalone product that finds every AI agent an enterprise is running across platforms, measures performance, and flags the riskiest ones. General availability is set for October.

SOFTWARE desk — a company that can count its servers still often cannot say how many AI agents it is running, who owns them, or whether they earn their cost. Dataiku’s announcement is a product that scans agents built on other companies’ platforms into one inventory and keeps a risk record for the ones that handle customers, sensitive data, or live transactions, with general availability set for October.

What the line above the lede says, and what the lede says the launch is. The line above the lede says Agent Management discovers, measures value, and flags risk for agents across your entire business, no matter where they were built. The lede says Dataiku, called the Platform for AI Success, today announced the launch of Agent Management, a standalone product that finds every AI agent an enterprise is running, regardless of which platform built it, measures the business and technical performance, and flags agents that pose the greatest risk. Unveiled at Dataiku Succeed, the company’s annual flagship conference, this new standalone product will be generally available in October. An AI agent, here, is software that can take the next step on a task, not only answer one question. A standalone product, here, means it is sold on its own, not only as a switch inside another Dataiku screen. Those glosses are this desk’s. The release prints the words. Today announced, and new, are the release’s. The release does not print a first-ship date past that wording. The lede says generally available in October. The availability section says October 2026. They are the same claim, worded twice. Do not file them as two dates. This filing does not treat 24 Sep 2026 as a first-ever product date the page does not print.

Why the release says companies cannot see the agents, and whose figure that is. The launch addresses a disconnect between how fast agents are being created and adopted and organizations’ ability to observe and manage them. Every large company already keeps an account of the software it runs: who owns it, what it costs, and when it renews. Almost none can say the same about the AI agents. Fewer than one in five organizations maintain a complete, current inventory of their AI systems, according to IBM’s “AI in Motion” research. One in five is 20 percent. Fewer than one in five is less than that, if the citation holds. That translation is this desk’s. The fraction is the release’s citation of IBM. It is not a count this desk made. The reason, the release says, is that most agent platforms can only see the agents built on them, leaving the bulk of a company’s agent estate without clear ownership, purpose, risk, or business outcome. An agent estate, here, is the set of agents a company is running. That gloss is this desk’s. The release prints the words. The link on the release for that IBM citation goes to the IBM Institute for Business Value report page whose path is /report/ai-orchestration-layer. The page title this desk read is “AI in motion,” and the page says it is in collaboration with the Dubai Future Foundation, on orchestrating AI governance. A question on that page, “Why is orchestration essential now?”, answers that only 18% maintain a current AI inventory. Eighteen percent is less than one in five. That translation is this desk’s. The 18% is that page’s. The release’s words are fewer than one in five, and complete, current inventory of their AI systems. The IBM page this desk read does not print that sentence. Do not collapse them. This desk did not audit IBM’s sample.

Florian Douetteau, as a quote, not as a count. Florian Douetteau, co-founder and CEO of Dataiku, said ask a bank how many servers it runs, and you get an answer to the decimal. Ask how many AI agents it is running, and you get a shrug or a guess. He said nobody set out to build it this way. Teams built agents faster than anyone could count them. Agent Management tells you what is actually out there, and what it is actually worth. To the decimal, and a shrug or a guess, are his sentences. They are not a server count or an agent count this desk made. What it is actually worth is his sentence. It is not a dollar figure on the release. Those are his sentences on the release. This desk did not interview him.

Which platforms the release names, and what the scan is supposed to return. Agent Management connects to the platforms that enterprise teams already use to run agents, including AWS Bedrock, Databricks Agents, Google Vertex, Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce, Snowflake Cortex, Dataiku itself, with OpenTelemetry support for custom environments, and scans them into a single inventory. Bedrock is Amazon’s service for running AI models. Vertex is Google’s. Copilot Studio is Microsoft’s tool for building agents. Azure Foundry is the other Microsoft platform the release names in that same phrase. Agentforce is Salesforce’s. Cortex is Snowflake’s. OpenTelemetry is a shared way for software to report what it is doing, so another tool can watch programs it did not build. Those glosses are this desk’s. The release prints the names. Including means the list is the release’s examples. It is not a sentence that these are the only platforms. It automatically identifies each agent’s structure, including the tools and models it relies on, giving supervisors full transparency into how an agent actually works, not just that it exists. Full transparency is the release’s claim. This desk did not open an inventory.

What the release says it keeps for the riskiest agents, and only those. For the highest-risk agents, the ones handling customers, sensitive data, or live transactions, it keeps a standing record of certification status, named risks, and tests that rerun on a schedule, so the evidence trail already exists when a manager, auditor, or regulator asks for it. Certification status, here, means whether the agent has been cleared. A named risk is a risk written down, not only a score. A scheduled retest is a check that runs again later. Those glosses are this desk’s. The release prints the words. Highest-risk, in that sentence, is the release’s category for agents that handle customers, sensitive data, or live transactions. This desk did not certify an agent.

What the release says a single platform cannot answer. Unlike traditional agent monitoring, which the release says sits in a single vendor’s stack and is designed to favor that vendor’s own agents, Agent Management is deliberately agnostic. It sits above the stack, which lets it answer questions no single platform can: which agents are unmonitored, where risk is concentrated, and which ones earn their cost. Teams ask those questions in plain language and get an answer covering the whole portfolio. Agnostic, here, means it is not built to prefer one vendor’s agents. Plain language means a sentence a person would say. A portfolio, here, is the whole set of agents, not a stock portfolio. Those glosses are this desk’s. Deliberately agnostic, and earn their cost, are the release’s words. They are not a bill this desk checked. This desk did not ask the product a question.

When it ships, what the price line says, and who the about box says the company is. Agent Management will be generally available in October 2026. Product access is priced per instance annually, with monitoring metered per agent. An instance, here, is one installation of the product. Metered per agent means the monitoring charge follows how many agents are watched. Those glosses are this desk’s. The release prints the words and does not print a dollar amount. The for-more line is https://www.dataiku.com/blog/dataiku-agent-management. The about box says Dataiku is the Platform for AI Success, the enterprise orchestration layer for building, deploying, and governing AI. In a single environment, teams design and operate analytics, machine learning, and AI agents. Sitting above data platforms, cloud infrastructure, enterprise applications, and AI services, Dataiku connects the full enterprise AI stack. The world’s leading companies, and a true business performance engine, are the about box’s words. This desk did not rank vendors. The media contact is Gina King, press@dataiku.com. This desk did not email that address.

What the card shows, and what the 24 Sep release does not print. The card is the Agent Management product UI from the product page at dataiku.com/product/agent-management, a headless crop. The still is an inventory table. The rows this desk can name are Incident Root-Cause Agent on Amazon Bedrock, Plant Onboarding Assistant, Defective Parts Return Agent on Databricks, and Predictive Maintenance Agent on Dataiku. The caption on the still includes GOVER AT SCALE. Those names and that caption are the product UI’s. The 24 Sep release does not print them. The schema image on the FinancialContent page is a site image at cdn.newsrpm.com, not this table. This filing does not use that image. The card has no date text.

What the product page adds, and what it does not let you file as the wire. The page, read 24 Sep 2026, says Agent Management is a standalone product that gives an organization one place to discover, monitor, and manage the risk of all its AI agents, wherever they were built. The platform answer on the page lists Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce, AWS Bedrock, Google Vertex, Databricks, Snowflake Cortex, n8n, and Dataiku itself. n8n is that page’s extra name. The wire says Databricks Agents and does not print n8n. The wire’s custom-environment line is OpenTelemetry. This page’s platform answer does not print OpenTelemetry. Do not collapse the two lists. The page says it is not an AI gateway. A gateway sits between agents and their tools and controls traffic. Agent Management sits above the agents and keeps track of them. For agents built in Dataiku, the LLM Mesh already plays the gateway role, and Agent Management sits on top of that too. An LLM is a large language model, the kind of model that writes and answers in sentences. A mesh, here, is Dataiku’s layer for those models. Those glosses are this desk’s. The page prints LLM Mesh. The wire does not. Availability on this page: October 2026, with a request-a-demo link. Connection, page: native APIs and log streams; a person selects the framework, authenticates, and the system scans; the depth of integration depends on what each platform exposes, with deeper integrations actively being developed. Actively being developed means those deeper links are not described as finished. That gloss is this desk’s. The page also says Dataiku was named a Leader in the 2026 Gartner Magic Quadrant for AI Platforms for Data Science and ML, for the fifth consecutive year. That line is this page’s. It is not on the 24 Sep release this desk read. This desk did not request a demo, and it did not read the Gartner report.

What the blog the release names adds, and what it is not. The page at dataiku.com/blog/dataiku-agent-management is titled “Discover Dataiku Agent Management: every agent on the record,” dated September 24, 2026, 4 min read, by Julia Tran. The platform sentence on that page lists Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce, AWS Bedrock, Google Gemini Enterprise, Databricks, Snowflake Cortex, or even coded agents on Open Telemetry. Google Gemini Enterprise is that page’s Google name. The wire’s Google name is Google Vertex. Open Telemetry, two words, is that page’s spelling. The wire prints OpenTelemetry as one word. Do not collapse Gemini Enterprise and Vertex, and do not collapse the two spellings into a second product. The blog says Agent Management does not operate your agents or step into their decisions, so they keep running exactly as your teams built them. That sentence is the blog’s. The wire does not print it. The blog’s line about an agent with three users and an agent with 3,000 is an illustration. It is not a customer count. This desk did not count users.

Plain English for the rest of the card. Agent Management is the inventory. It is supposed to find agents built on other companies’ platforms, measure how they perform, and flag the risky ones. Generally available in October, and October 2026, are the release’s two wordings of the same ship window. Per instance annually, and metered per agent, are the price shape. They are not a dollar price. Fewer than one in five is the release’s citation of IBM. 18% is the IBM page’s separate line for a current AI inventory. 9:00 AM EDT is 9:00 a.m. Eastern and 1:00 p.m. UTC. 13:00:43 UTC is the syndication’s modified clock, forty-three seconds later. Bedrock, Databricks, Vertex, Copilot Studio, Azure Foundry, Agentforce, Cortex, and Dataiku are the names on the wire. n8n is the product page’s extra name. Gemini Enterprise is the blog’s Google name. OpenTelemetry is the wire’s name for custom environments. Florian Douetteau is co-founder and chief executive. Gina King is the media contact. Julia Tran is the blog byline. Incident Root-Cause Agent, Plant Onboarding Assistant, Defective Parts Return Agent, and Predictive Maintenance Agent are the still’s. This filing is the 24 Sep announcement.

PRIMARY here: Dataiku’s 24 Sep 2026 Business Wire, release 20260924350178, stamped September 24th 2026, 9:00 AM EDT on the FinancialContent syndication this desk read — Tier A PRIMARY, the company’s own announcement. STATUS PRIMARY. businesswire.com itself did not return a page. The launch at Dataiku Succeed, the standalone-product sentence, generally available in October and October 2026, the fewer-than-one-in-five citation of IBM’s “AI in Motion,” the platform-blind-spot sentence, the Douetteau quote, the platform list including Bedrock, Databricks Agents, Vertex, Copilot Studio and Azure Foundry, Agentforce, Snowflake Cortex, Dataiku, and OpenTelemetry, the structure sentence, the highest-risk certification sentence, the agnostic and earn-their-cost sentences, the per-instance and per-agent price shape, the about box, and press@dataiku.com are that release’s. The 18% current-AI-inventory line is the IBM report page’s. n8n, the not-a-gateway sentence, LLM Mesh, and the Gartner Magic Quadrant line are the product page’s. Gemini Enterprise, the two-word Open Telemetry spelling, the does-not-operate sentence, and the three-users and 3,000 illustration are the blog’s. The four agent names and GOVER AT SCALE are the product UI’s. NOT claimed: that this desk opened businesswire.com, that fewer than one in five and 18% are the same sentence, that either figure was counted here, that the platform lists are identical, a dollar price, that the four named agents are customers this desk verified, a login, a stock tip, or investment advice. The card is the Agent Management inventory UI. Distinct from the already-filed gurucul-ai-risk-response, avalara-versori, and autonomyai-discover-mode.