The new operating system for communication

Whitepaper on Agentic AI for free download

Agentic AI is not the next AI feature. It is the question of who will still be shaping communication in five years’ time – and who will only be executing it. In a sprint group with 14 experts, the CommTech working group analyzed where the technology stands, what it means for teams – and where the real pitfalls lie.

2030: Two scenarios, one course set

Scenario A sounds tempting: the “Operating System of Comms”. Agents take over monitoring, briefings and repurposing. In the morning, the media draft is ready – with evaluation, escalation signal, preliminary response for release. The press spokesperson makes the decision and the system prepares everything else. Microsoft proves that 16 out of 20 steps of an earned media story can already be automated today. Only steps 5 to 8 – story idea, target group, pitching, spokesperson briefing – remain decidedly human. After all, we want to appeal to people’s tastes and cultivate relationships with them.

Scenario B is the opposite: the “hive mind”. An agent picks up on a false report, responds automatically – no one has proofread. LLMs confirm incorrect premises in studies by up to 90 percent. Whoever controls the model controls the communication logic. Dario Amodei, CEO of Anthropic, has put it clearly: Agentic systems bring enormous productivity gains and at the same time new risks of mismanagement, abuse and concentration of power. That is not a disclaimer. That is the specification.

8 percent are ahead – 71 percent still before the decision

The CommTech Index Report 2025/2026 shows: 88% of communication departments are experimenting with AI, 51% have integrated tools. But only 8 percent are using agentic AI productively. The window for early movers is open. Waiting is not a neutral option, it is a decision for the status quo.

The biggest scaling hurdles according to Celonis (2026): Silos between teams and systems (54%), lack of AI expertise (47%), IT-business alignment (45%), security and governance (43%). Resistance to change? Statistically at 6 percent. What slows things down is not refusal – it is inability to operate. And our own observation from the sprint group: communication is too often reactive. Marketing is faster. IT takes the lead. If you don’t actively forge alliances, you become an execution unit.

From prompt to agent: What Agentic AI really is

A chatbot responds to prompts. One step, then it waits. An AI agent reacts to trigger events, performs multi-step tasks and finds its own solutions. Agentic AI goes even further: the system develops strategies independently and adapts actions dynamically – decision paths do not have to be pre-programmed.

Two concepts are technically crucial: Function calling enables the AI to trigger external APIs and database queries – instead of guessing, it retrieves real data. This reduces hallucinations and raises quality to series production level. The Model Context Protocol (MCP) – introduced by Anthropic as an open standard – is the “USB-C connection” for this: a standardized interface between AI and internal and external company systems (CMS, CRM, media monitoring, intranet). The AI acts in the real working world – safely, because MCP defines exactly what it is allowed to see or process.

Four new roles – and one unchanged

When agents move in, who does what in the team changes. The sprint group identified four key roles: the Agent Orchestrator (strategic coordination of agent instances), the Agent Builder (construction of workflows and tool connections), the Agent Ops (system stability and monitoring) – and the Human in the Loop: final authority for quality, tonality and ethical integrity. This role has not been left behind. It is the center.

What Microsoft® and YessPress® are already building today

André Pechmann, Director Communications at Microsoft® Germany, sees the decisive trend of the next 18 months in the transition from assistive to autonomous AI. Using the Microsoft Graph – with access to Outlook, Teams, SharePoint and CRM – AI recognizes complex relationships across departmental boundaries and executes actions, not just suggestions. Pechmann’s practical tip: use agents as stress testers. The agent plays the skeptical journalist, the competitor, the target group – and uncovers gaps in the briefing before they appear in the interview.

Thomas Massmann from YessPress® has already priced in the end of watering can communication: In five to ten years, machine-readable, individualized communication will replace the classic press distribution list. And he has built it: an Agentic AI module for automated responses to media inquiries, which is currently going live at a large corporation. Function Calls combine IT and AI; an embedding engine qualifies journalists’ questions and the language model generates high-quality answers. The press officer sees the suggested answer even before she has read the question. Then she decides and presses send. This is “Human in the Loop”.

Decision Matrix: Not everything that works should become a use case

The Sprint Group has developed an evaluation framework that classifies use cases according to value (operational value contribution + communication impact) and effort (feasibility + operation) – in four quadrants: Quick Wins (start immediately), Strategic Projects (prepare), Gap Fillers (at capacity) and Time Wasters (avoid).

All scores are preceded by the “red line”: GDPR conformity, compliance, information security, ethics. If a governance criterion is violated, the use case is rejected – regardless of the score. The framework is available for download in the free AG CommTech whitepaper.

Five recommendations for getting started

– Use the Decision Matrix now. Evaluate current and planned AI projects according to value and effort. Sorting out is part of this.

– Build a quick-win agent. The daily media briefing (preferably on the good database of a service provider with an MCP interface) is ideal: clearly defined scope, measurable ROI (30-60 minutes saved daily), manageable data connection.

– Governance before technology. Clarify first: What is AI allowed to see? Who has the final say? How are outputs documented?

– Forge alliances now. Communication must be a co-creator in AI projects, not a late adopter. If you wait until IT is ready, you will get what others have imagined.

– Rethinking roles today. Which person in the team could become an orchestrator? Which builder? The window of opportunity for early movers is still open.

“In the end, perhaps 80 percent of processes will be covered digitally – while 20 percent will be covered by the human-inthe-loop will remain.”

– Thomas Massmann, YessPress (2026)

About the author: Richard Tigges is Director of the Institute for Applied Communication Intelligence and is Co-Lead of the AG CommTech Cluster Technology. Martin Regnet is a senior expert in digital corporate and financial communications and led the AG CommTech sprint group “Agentic AI”, which wrote the white paper.