3 in 4 Companies Are Planning AI Agents: What Does That Mean for Your Business If You're Still Hesitating?

A Friday afternoon in July, and another notification lands in your inbox: Deloitte has just published the report "The State of AI in the Enterprise: The Untapped Edge." The headline finding that's since been making the rounds through trade media and LinkedIn feeds: nearly three in four companies worldwide plan to deploy agentic AI within the next two years. Maybe you've already seen the number somewhere. Maybe you noted it briefly and moved on to the next email. That's exactly what we want to stop you from doing in this article – because there's more behind the headline than a single percentage suggests, and for your business, the actually interesting question is a different one: what does this development concretely mean if you yourself are still hesitating?

We put the numbers in context, show the flip side of the ambition, explain why the new large-enterprise packages from Accenture and Google Cloud bypass the classic German Mittelstand, and say quite openly where things actually stand in Germany. In the end, there's no panic-mongering, just a sober assessment – and a few concrete next steps.

74 percent: the number you shouldn't ignore

The Deloitte report is based on a survey of 3,235 business and IT executives – from director level up to the C-suite – across 24 countries, conducted between August and September 2025. This isn't a niche survey among tech enthusiasts, but a broad snapshot of what decision-makers at large and mid-sized organizations are currently planning. Deloitte itself speaks of "close to three-quarters" or "nearly 3 in 4" companies planning to deploy agentic AI at least moderately within the next two years. Several independent secondary sources consistently put this share at 74 percent – of which 23 percent plan extensive deployment, and 5 percent want to establish AI agents as a fully integrated core part of their operations.

For context: the exact figure of 74 percent doesn't appear in the official Deloitte press release as a single, isolated metric, but is phrased there more descriptively. Still, the fact that several independent sources arrive at the same value independently of each other makes the number fairly reliable – we use it here accordingly, and transparently.

Even more important than the raw percentage is a second finding from the same report: 85 percent of surveyed companies expect they will need to customize AI agents to their specific business requirements. An off-the-shelf agent isn't enough for most of them. That's a remarkable statement, because it contradicts the picture many vendors are currently painting – namely that agentic AI can be installed like an app from a store. The reality that emerges from the Deloitte survey is different: anyone who wants to deploy AI agents meaningfully has to adapt them to their own processes, data, and decision paths. That matches our own experience from practice exactly – generic solutions rarely fail because of the technology, but because they don't fit actual day-to-day work.

The flip side of the ambition: governance falls behind

If you now think that 74 percent readiness to deploy is automatically a reason for concern because "everyone else" is already so far ahead, you should know the second central finding of the Deloitte report. It puts the sense of momentum into significant perspective: only one in five companies – 21 percent – has, according to Deloitte, a mature governance model for autonomous AI agents. So the gap between ambition and actual control is large, in many cases even very large.

This gap also shows up in the biggest concerns executives named in the survey: 73 percent see data protection and security as the main risk, 50 percent name legal questions around IP and regulation, and 46 percent each worry about governance capabilities and about model quality and explainability. In other words: the companies planning to deploy right now know very well themselves that they don't yet have the control mechanisms under control. They're planning anyway – but not blindly.

Nearly three in four companies plan to deploy agentic AI at least moderately within the next two years. Only one in five already has a mature governance model for it. It's exactly in this gap between ambition and control that it's decided who benefits from AI agents – and who buys themselves problems.

For your business, that means: hesitating isn't inherently wrong. But blindly diving in without a governance concept would be just as poor an idea as waiting forever. The real question isn't "When do I get in?" but "With which use case, which data guardrails, and which accountability do I get in?" Anyone who has engaged with the EU AI Act already knows this way of thinking – we've summarized the concrete obligations for the German Mittelstand in a separate article on the EU AI Act.

Why the new enterprise packages bypass the German Mittelstand

On July 7, 2026 – barely a week before this article was published – Accenture, through its newly founded unit "Accenture Edge," and Google Cloud announced a joint suite of pre-built agentic AI solutions for the so-called mid-market. At first glance, that sounds like exactly the answer many companies are looking for: ready-made packages instead of months of in-house development, deployment "in weeks instead of months," covering six areas from customer intelligence through cybersecurity to agentic business processes and workforce enablement, powered by the Gemini Enterprise App, the Gemini Enterprise Agent Platform, and the Agentic Data Cloud, backed by Accenture's "Forward Deployed Engineers."

The second look, though, is worth it. The target group for these packages is explicitly companies with annual revenue between $300 million and $3 billion. That's a scale that has little to do with what's commonly considered the Mittelstand in Germany. Under the IfM Bonn's definition, companies with 10 to 499 employees and annual revenue between €1 million and €50 million count as Mittelstand here – there's no legal definition, but this classification is the established reference framework. Even at the upper end of that range, a typical German mid-sized company sits far below the revenue floor that Accenture Edge and Google Cloud have set for their new offering.

It's also worth noting what's missing from the official press release of July 7, 2026: neither Germany, nor the DACH region, nor Europe is mentioned there. The offering is visibly tailored to the US or global mid-market in the stated revenue class – not to a mid-sized company with 40, 80, or 150 employees somewhere between Munich and Hamburg.

That's not a footnote, but the actual point of the current news. The big providers are visibly building offerings for large enterprises and large mid-market companies in the US sense right now – not for the classic German Mittelstand. Anyone who believes "the big players will take care of it, I'll just wait for a ready-made product for my size" will likely wait a long time. At the same time, Deloitte's 85 percent figure shows that even large corporations have to customize their agents individually – so a generic product is only ever a starting point, never a finished solution.

Where Germany actually stands: the Bitkom 2026 numbers

So how far along is the German Mittelstand really? The Bitkom AI Study 2026, a representative survey of 604 German companies with 20 or more employees, provides solid numbers on this. According to it, 41 percent of German companies now actively use AI – up from just 17 percent in 2024. That's more than a doubling within two years. Another 48 percent plan to adopt it. Taken together, nearly 90 percent of surveyed companies are already actively engaging with AI or have firmly planned to.

The decisive question, though, is why a significant share of these companies aren't as far along as their own plans actually call for. The Bitkom study names three main reasons: lack of AI expertise on the team (53 percent), uncertainty around data protection (41 percent), and unclear costs (37 percent). That aligns almost one-to-one with the concerns also named in the global Deloitte survey – data protection and security top that list too, at 73 percent. The difference between a company that's "hesitating" and one that's "planning" is thus often not a lack of will, but a lack of expertise, unresolved data-protection questions, and uncertainty about actual costs.

This is exactly where the real difference between a generic enterprise rollout and tailored guidance lies: a smaller, specialized agency can address exactly these three points – building expertise, clarifying data protection, transparent costs – while a package like the one from Accenture Edge and Google Cloud is designed for an entirely different company size and a different market. We've already described in detail what it means when teams use AI tools but don't yet have a real strategy behind them – for instance in our article on why using ChatGPT alone still isn't an AI strategy.

Important for context: the Bitkom sample only covers companies with 20 or more employees. Smaller businesses, which make up a large share of the classic German Mittelstand, aren't represented in it – so the real gap among very small companies is likely even bigger than the numbers suggest.

Don't dive in blindly: what Gartner warns

For all the justified urgency, it would still be wrong to now rush headlong into any AI agent project just to be part of it. Gartner already predicted back in June 2025 – and the forecast is still widely cited in 2026 – that more than 40 percent of agentic AI projects will be abandoned by the end of 2027. Gartner cites rising costs, unclear business value, and inadequate risk controls as reasons. For scale: development costs alone for individual agents, say in HR or customer service, are put at $40,000 to $150,000.

This figure is explicitly a forecast for a future point in time, not an already-measured, realized outcome – worth keeping in mind when weighing it. Still, it's an important counterweight to the sheer momentum of the Deloitte numbers. It shows: "act now" must not mean "do anything with AI agents, as long as it's fast." Projects that fail generally don't do so because the underlying technology doesn't work, but because the business case was unclear from the start or governance was missing – the same gap Deloitte also finds in roughly four out of five companies that lack a mature governance model.

For your business, that means very practically: a single, clearly scoped use case with measurable benefit almost always beats an ambitious large-scale project without a clear business case. We show how quickly such a focused agent can actually become productive in our piece on how a Mittelstand AI agent can go live in hours, not months – without any of the complexity of a months-long enterprise rollout.

What this means for your business

Let's sum up the situation without dramatizing it: according to Deloitte, nearly three in four companies worldwide plan to enter agentic AI at least moderately within two years. That figure comes from a global enterprise survey, not specifically from the German Mittelstand – transferring it one-to-one would be irresponsible. At the same time, the Bitkom numbers show that in Germany too, 41 percent of companies already actively use AI and another 48 percent plan to. The real lag many German businesses face is rarely a lack of will, but a lack of expertise, unresolved data protection, and unclear costs.

At the same time, as the example of Accenture Edge and Google Cloud shows, the big providers are building their newest packages for a company size that has nothing to do with the classic German Mittelstand. Anyone waiting for a ready-made, affordable enterprise solution "off the shelf" that happens to also fit 80 or 150 employees will likely be disappointed. And the Gartner forecast is a reminder that even among companies that are already getting started, more than one in three projects could fail due to an unclear business case or missing governance.

What follows: hesitating costs time that competitors are currently using to build expertise. But diving in blindly without a clear use case, without clarifying data protection questions, and without a realistic cost picture is just as risky. The middle path – and this isn't a convenient excuse, but the consequence of every study cited here – is a focused, individually tailored first use case, accompanied by someone who keeps both the technical implementation and the governance questions in view. That's exactly the difference between an enterprise template built for billion-dollar revenues and a tailored solution for your business.

The 74 percent from the Deloitte report is therefore less a threat than an invitation to ask yourself the right questions now – with a clear head, not with last-minute panic: which process in your company costs the most time and can be clearly scoped? Which data may even be processed in the process? And who in the company takes responsibility once the agent is running? Anyone who can answer these three questions is already well ahead of most of the 3,235 executives surveyed, who are planning but don't yet have a governance model.