Agentic AI · BraivIQ AI Blog

Agent Management Becomes A Product Category: Why 62% Of Companies Are Experimenting With AI Agents But Fewer Than 10% Have Scaled One

On 24 September 2026, Dataiku announced Agent Management - a standalone product whose entire job is to keep track of the AI agents a company already has: inventorying them across platforms, tracking their business KPIs and technical performance, and tiering them by risk, with general availability planned for October. A product like that only exists because of a problem, and the problem has a hard number attached: McKinsey finds that 62% of organisations are experimenting with AI agents, but fewer than 10% have scaled them in any function. The missing layer, it turns out, is not another agent - it is management of the ones you have. Most businesses cannot say what agents are running, where, what each is for, whether it is delivering, or how risky it is, and without those answers an agent stays a pilot forever. This analysis explains what agent management means in plain terms - inventory, KPIs, risk tiers, ownership - why it is the bridge from pilot to production, and how any business can apply the discipline right now, whether or not it ever buys a product for it.

September 24, 2026 · 11 min read · By BraivIQ Editorial

24 Sept - 2026 - Dataiku announces Agent Management: inventory, KPIs, technical performance and risk tiers for agents across platforms · 62% vs <10% - Organisations experimenting with AI agents versus those that have scaled them in any function (McKinsey) · Not another agent - The missing layer is management of the agents you already have, not a new one · GA in October - A standalone product for agent management is arriving because the category now exists

A new kind of product tells you something about the state of an industry, and on 24 September 2026 a telling one arrived. Dataiku announced Agent Management: a standalone product whose whole purpose is to keep track of the AI agents a company already has - inventorying them across the platforms they run on, tracking their business KPIs alongside their technical performance, and tiering each agent by risk - with general availability planned for October. Nobody builds a product to manage something unless enough of it exists to be unmanageable, and the number behind this one is stark: McKinsey finds that 62% of organisations are experimenting with AI agents, yet fewer than 10% have scaled them in any function. That gap is the central puzzle of enterprise AI this year, and a management product for agents is the industry's answer to it. The missing layer between pilot and production is not a better agent, a cleverer model or one more use case - it is the ability to know what agents you have, what each is for, whether it is delivering, and how risky it is. As an AI Agency London that helps businesses move agents from experiment to operation, we think agent management is the least glamorous and most necessary idea of the season, and this analysis is what it means and how to apply it.

Why Experiments Do Not Become Operations: The path from 62% experimenting to under 10% scaling is not mysterious. An agent pilot starts in one team, does something useful, and then stalls, because scaling it means answering questions the pilot never had to: who owns this agent, what is it allowed to do, how do we know it is working, what does it cost, what happens when it goes wrong, and how does it fit with the other agents other teams have quietly built? Multiply that by a dozen pilots across a business - each on a different platform, each with a different owner or none - and you have agent sprawl without agent knowledge: plenty of agents, no picture of them. Leadership will not, and should not, scale what it cannot see and measure, so the pilots stay pilots. Agent management is the answer to precisely this: the inventory, measurement and risk classification that let a business say, with confidence, which agents deserve to scale.

What Agent Management Actually Means

Strip away the product branding and agent management is four disciplines that any business can understand and most have never applied to AI. Inventory: a single register of every agent running in the business - across every platform and vendor, including the ones a team built without telling anyone - with what it does, which systems it touches, and who owns it. You cannot manage, secure or scale what you have not counted, and most organisations have not counted. KPIs: each agent measured against the business outcome it exists to deliver - tickets resolved, invoices processed, leads qualified, hours saved - as well as its technical health, so that 'is it working?' has an answer in numbers rather than anecdotes. Risk tiers: agents classified by what they can do and what could go wrong - an agent that drafts internal summaries is low-risk; one that can move money, contact customers or change records is high-risk - with the governance, approvals and monitoring proportionate to the tier, so that oversight is concentrated where consequences are. And ownership and lifecycle: a named owner for each agent, and a defined path from pilot through production to retirement, so agents are decided about rather than accumulated. A product like Dataiku's packages these into software; the disciplines themselves are simply good management applied to a new kind of worker.

Inventory - one register of every agent, across platforms, with its purpose, the systems it touches and its owner.
KPIs - each agent measured against the business outcome it exists for, plus its technical health, in numbers.
Risk tiers - agents classified by capability and consequence, with governance and monitoring proportionate to the tier.
Ownership - a named person accountable for each agent's performance, behaviour and cost.
Lifecycle - a defined path from pilot to production to retirement, so agents are decided about rather than left to accumulate.

Why This Is The Bridge From Pilot To Production

Agent management is not bureaucracy layered on innovation; it is the thing that makes scaling an agent a rational decision rather than a leap of faith. With an inventory, leadership can see the whole estate and stop duplicating effort. With KPIs, the pilot that genuinely delivers is distinguishable from the one that merely impresses, and the case for scaling it is made in the currency the business already speaks - outcomes and cost - rather than in enthusiasm. With risk tiers, the low-risk agents can be scaled quickly with light oversight while the high-risk ones get the controls they warrant, instead of every agent being held to the same paralysing standard or, worse, none. With ownership, someone is accountable for taking the agent from pilot to production and keeping it healthy there. This is exactly the machinery a business uses to scale anything else - a product, a process, a supplier - and its absence for agents is why they have piled up as experiments. It is also, not coincidentally, the discipline that lets a business govern the agents it now buys from its software vendors as well as the ones it builds, because an inventory and a risk tier apply equally to an agent that came bundled with the CRM. The businesses that will be in the under-10% next year are the ones building this layer now, with or without a product.

The Bottom Line

Dataiku's Agent Management, announced on 24 September and arriving in October, exists because the industry has hit a wall that McKinsey measures precisely: 62% of organisations are experimenting with AI agents and fewer than 10% have scaled one, and the reason is not a shortage of agents but an absence of management - most businesses cannot say what agents they run, what each is for, whether it delivers, or how risky it is, so their pilots stay pilots. Agent management is four plain disciplines: an inventory of every agent across platforms, KPIs that measure each against its business outcome, risk tiers that concentrate governance where consequences live, and named ownership with a lifecycle from pilot to production to retirement. It is the bridge from experiment to operation because it turns scaling an agent into a rational, evidenced decision made in the currency of outcomes and cost, and it governs bought agents as well as built ones. Any business can start applying it now - make the inventory, define the outcome, classify the risk - and discover which pilots deserve to scale. The businesses in the under-10% next year will be the ones that built this layer, and helping businesses build it, then take the agents that earn it into production, is exactly the work we do.

References & Further Reading

Solutions Review - top worktech news from the week of September 25th (Dataiku Agent Management announcement): https://solutionsreview.com/enterprise-resource-planning/top-worktech-news-from-the-week-of-september-25th-updates/
Dataiku - newsroom and product announcements: https://www.dataiku.com/company/news/
Omago - SME AI adoption in 2026: what the data actually shows (McKinsey: 62% experimenting, under 10% scaled): https://www.omago.ai/blog/sme-ai-adoption-2026-data
AI to ROI - news and analysis, September 25, 2026: https://ai2roi.substack.com/p/ai-to-roi-news-and-analysis-september-780
McKinsey - the state of AI: agents, innovation and transformation: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai