Key findings

• 98% of companies have deployed AI agents or plan to: 44% already in production (mostly within limited functional areas), 40% experimenting, 14% planning 

  
  

• 48% have no clearly defined responsibility for AI agent management, with multiple departments involved instead of a single owner 

  
  

• Enterprise architects are involved in AI agent management in only 34% of companies, and hold sole ownership in just 5% 

  
  

• 63% of enterprise architects believe their own function should be driving strategic adoption and governance 

  
  

• More than 80% of companies lack visibility into how deployed agents perform, whether they are compliant, or what ROI they deliver 

When AI agent adoption is done, management becomes the issue. 

The question that matters right now is not whether companies will use AI agents. It is who will be accountable for them once they are running.  
 
The SAP LeanIX Agentic AI Survey 2026, based on more than 200 enterprise architecture leaders across Europe and North America surveyed in March 2026, makes the scale of adoption clear: 44% of companies have already deployed AI agents, mostly within limited functional areas, another 40% are experimenting, and 14% are planning to. 

Ninety-eight percent of respondents fall into one of these three groups. So it's fair to say that when the adoption phase is settled, management becomes the real issue to handle. 

What is an enterprise architect? 
An enterprise architect is the professional responsible for aligning an organization's IT systems with its business strategy. The role sits at the top of the technical hierarchy within IT, combining deep technical expertise with a business-level view. 

How do you know if your AI agent is really performing well? 

When preparing to implement AI agents within a company, one question above all others guides your thinking, before, during, and after the process: how can I ensure that artificial intelligence delivers tangible, relevant results for my work and that of my team? 

Per the survey, nearly half of respondents agree that AI initiatives need to show tangible results to get approved in the first place. What stands in the way is not the technology itself: only 15% point to limitations like hallucinations as a major obstacle. The two more structural gaps are lack of employee know-how, cited by 58% as a primary or secondary barrier, and difficulty identifying which use cases are worth pursuing, cited by 56%. 
  

Even where AI agents are already running, most organizations cannot see what they are doing. Roughly half of enterprise architects report at least partial visibility into which agents are in use. That visibility collapses further down the funnel: more than 80% lack insight into how deployed agents support business capabilities, how they perform, whether they are compliant, or what return they deliver. More strikingly, 38% of companies say they have no plan to establish that visibility in the future. 

So while we hear the most about RAG and Knowledge Base or Harnessing, visibility seems to be one of the most important words to take into consideration when we speak about AI agents and all the investments around it. 
  

Who is in charge of AI agent governance? 

Governance is considered so necessary that nearly three-quarters of respondents agree AI agents need a dedicated framework, though most have not built one yet. 

To limit that risk, one of the strategies implemented is to restrict where and how agents can be deployed: 77% limit use at the department or business unit level, largely through mandatory review by a responsible body (72%) and restrictions to approved vendors or platforms (52%) 
  

What none of that guardrail-building solves is who is actually in charge. Only 37% of companies have a clearly designated team responsible for AI agent management. In 48% of cases, responsibility is split across multiple departments with no single owner, and satisfaction tracks that ambiguity closely: 81% of respondents are satisfied when one team owns the process, against just 7% when nobody does. 

The Role of Enterprise Architects in AI Implementation. 

Enterprise architects and the way they are engaged in the process illustrate this issue better than any other example. According to the survey, the function is involved in AI agent management in only 34% of organizations, and holds sole ownership in just 5%, well behind IT teams more broadly. This is despite the fact that 63% of enterprise architects believe their own discipline should be driving strategic adoption and governance. 

The data suggests what’s behind this situation. Enterprise architects claim ownership readily where the work resembles what they already do, but their confidence drops sharply once the question shifts to an agent's ongoing lifecycle: only 27% think the enterprise architecture function should own agent performance and ROI, the largest gap found in this comparison.  

The data that sums it all up? Seventy percent of enterprise architects agree that agentic AI creates unique challenges for their practice and that new tools and methods are needed to address them. 

SME and EA: Two Sides of the Same Coin 

The instinct to add another framework or another review step will not close this gap. What the data points to instead is a clearer split in who owns what. Subject matter experts are best placed to manage, within their own domain, the corrections, exceptions, and judgment calls no framework can fully anticipate. Enterprise architects are best placed to hold the wider view: how agents connect across business capabilities, where they overlap, where they create risk at the system level. 
  

Right now, most organizations have neither. They have agents in production, a governance framework on paper, and no one accountable for the distance between the two. 

This is the gap Syllotips is built to close. Every time an SME correction is captured and turned into governed memory, that answer becomes reusable across the organization, not a one-off fix buried in a single interaction. Across enterprise deployments, a single expert answer is reused between 14 and 50 times. That is the missing link between SME-level judgment and enterprise architecture-level visibility: correction by correction, the system builds exactly the cross-capability picture enterprise architects say they want but currently cannot see. 

Frequently Asked Questions 

What percentage of companies have deployed AI agents in 2026? 

According to the SAP LeanIX Agentic AI Survey 2026, 44% of companies have already deployed AI agents, mostly within limited functional areas, and a further 54% are either experimenting or planning to deploy them, bringing total adoption or intent to adopt to 98%. 

Who is responsible for AI agent governance inside most companies? 

In 37% of companies, a dedicated team owns AI agent management. In 48%, no single team is responsible, and multiple departments are involved instead. Enterprise architects are involved in only 34% of cases and hold sole ownership in just 5%. 

Do enterprise architects want to lead AI agent governance? 

Yes. 63% of enterprise architects surveyed believe their function should drive the strategic adoption and governance of AI agents, even though most are currently excluded from that role. 

What's the role of an SME (Subject Matter Expert)? 

A Subject Matter Expert is a professional who is up to date and has all the knowledge required to step in and correct an AI agent when something goes wrong, due to a lack of information or context. Through Syllotips, every SME correction is captured and turned into governed memory, reusable knowledge that closes the loop between what an expert knows and what the agent, and the rest of the organization, can draw on next.