AI as a Digital Employee: Why the Real-World Test Is Harder Than Expected
AI agents are being marketed as autonomous digital team members by Salesforce, SAP and others. Adoption is doubling — but 84 percent of companies have not adapted their roles and processes. What SMEs need to clarify before deploying.
Salesforce calls them Agentforce, SAP deploys Joule — both pursue the same promise: AI agents that autonomously take on tasks like a digital employee. 16.6 percent of mid-sized companies already deploy such autonomous AI systems — double the rate from a year ago, according to the Salesforce AI Index SME 2026. A further 37 percent plan to introduce them before the end of 2026.
That sounds like a clear trend. And the trend is real — but it conceals a structural gap that determines whether a deployment succeeds or fails from the outset. How AI integration and automation works in practice depends less on the technology than on the questions companies answer before the first deployment.
What AI agents actually do — and what they do not
An AI agent receives a goal and decides on its own which steps to take. It can categorise emails, compile proposals from CRM data, classify support requests — provided it has access to the right systems. The central finding of the State of AI Agents 2026: the biggest barrier is not the intelligence of the systems but secure and reliable access to production systems. That is precisely what causes pilots to fail in practice.
Three hurdles that decide the real-world test
- System integration (46%): The most common reason for failed pilots is not the AI itself but the missing or overly complex connection to existing systems — ERP, CRM, document management.
- Data quality (42%): AI agents are only as good as the data they access. 61 percent of German companies make little or no use of their data potential, according to Bitkom.
- Data protection and compliance (77%): Nearly four in five companies cite data protection requirements as a digitalisation hurdle. For AI agents that autonomously access sensitive business data, this question becomes even more acute.
The real gap: technology without process clarity
84 percent of companies deploying AI have not yet adapted their roles and processes to the new reality — so a Deloitte study from March 2026 finds. AI is added as an extra tool rather than used as a trigger to rethink workflows fundamentally. This is the classic failure pattern: the technology delivers, but the organisation stays the same. Structured IT consulting and technology assessment helps to take this step before the rollout — not once the pilot is already stalling.
A useful foundation for assessing which AI level fits which process is our post AI Levels Explained: From ChatGPT to AI Agents to Automated Workflows.
What SMEs should concretely check now
- Process before tool: Which specific, repeatable workflow should the AI take over — and is it described clearly enough that a new employee could learn it in an hour?
- Clarify data access: Which systems would the AI need to connect to? Are APIs available, and who is authorised to access which data?
- Data protection check first: Personal data, customer data, internal documents — which processes involve them, and what does that mean for AI deployment under GDPR?
- Pilot on a small scale: One clearly scoped process, four to eight weeks, a measurable target — then scale.
- Adapt roles: Who reviews AI outputs, who escalates errors, who maintains system access? Without answers to these questions, even technically functioning pilots still fail.
AI integration for SMEs — NoviCogi
This article was created with AI assistance and editorially reviewed.
