When Robert Whitehead stepped onto the Dublin stage at INMA’s Media Innovation Week, he carried both urgency and a sense of history. The INMA Digital Platform Initiative lead told news executives they were facing “the biggest change in your life.”

His subject was agentic AI — the next stage beyond generative AI — and his message was clear: Media leaders must lean in now, experiment fast, and prepare for disruption that will unfold at unprecedented speed.

To prove it, he didn’t just present the theory. He demonstrated. With the help of a volunteer plucked without notice from the audience, Whitehead tasked an AI system with building a live breaking news app for Dublin. 

“Call it The Indispensable,” he said. “Make it vibrant and visual, like a social media feed. Pull stories from The Irish Times, the Independent, BreakingNews.ie, RTÉ, Dublin Live. And you’ve got about 15 minutes.”

Just 28 minutes later, the app was live. On the big screen, delegates saw a polished, working prototype that just moments earlier had not existed.

“This is the difference between talking about transformation and experiencing it,” Whitehead said.

Defining agentic AI

Whitehead began with the distinction for media executives still catching up on the jargon. “Generative AI responds to prompts. Agentic AI pursues goals,” he explained.

Generative AI is reactive: It creates content when asked — an article, an image, a line of code. 

Agentic AI is proactive: It perceives its environment, sets sub-goals, and takes action. 

Whitehead leaned on the original academic definition dating back to 1997, when Stanley Franklin and Art Graesser described an autonomous agent as a system that not only perceives but “acts within [an environment] over time to pursue its own goals and influence the environment in a way that allows it to alter future perceptions.”

“This is not just content creation,” Whitehead told delegates. “This is outcomes. This is action. That is what makes it profound.”

He described three key tiers of agentic capability:

• Some agents are still tools, acting only when called. 

• Others operate as semi-autonomous co-pilots, drafting plans or suggesting actions while waiting for human approval. 

• The frontier is fully autonomous agents — systems that plan, adapt, and execute without waiting for permission. 

“That is where the biggest transformation lies,” he said.

Why it matters

To underline the stakes, Whitehead pointed to two voices outside journalism:

Amazon’s Swami Sivasubramanian has declared: “Agentic AI is going to be one of the biggest transformations in the technology industry ever.” 

Demis Hassabis, the Nobel prize-winning CEO of Google DeepMind, has gone further: “We’re looking at something 10 times bigger than the Industrial Revolution — and maybe 10 times faster.”

Whitehead reminded the audience that every general-purpose technology has re-engineered the media industry: steam automated printing, railways sped up distribution, electricity cut costs and led to an explosion of production which allowed price cuts that ushered in the world’s first mass medium, computers and telecommunications modernised production, the Internet lowered barriers to entry for all potential content creations. 

“Agentic AI will be on that list of general purpose technologies,” he said. “Only this time, the curve is exponential.”

The live demonstration

The live app build was more than a gimmick. It symbolised how agentic systems collapse production timelines. What once required weeks of development and collaboration can now be achieved in minutes by agents that orchestrate content feeds, formatting, and distribution autonomously.

Critically, the on-stage demo did not build upon a white-labelled app, but built a new product from the ground-up. He used Replit, the coding-as-a-service product, which has just had the fastest-growth of any SAAS business — a favourite among vibe-coders who only need to speak to it to create and re-shape it.

“We are moving from a world where we ask machines to do things to a world where machines decide what to do next on our behalf,” Whitehead said.

From table stakes to titans

Before looking forward, Whitehead grounded his audience in today’s reality. 

INMA’s initiative leads, Sonali Verma (Generative AI initiative) and Jodie Hoperton (Product & Tech Initiative) chipped in to help Whitehead create an updated list of what is regarded as AI table stakes in news media use.

INMA then asked 100 companies to rank their adoption of these about their AI use, and the responses showed most publishers were already experimenting across a wide set of functions. 

“These are the basics,” Whitehead said. “If you’re not doing them yet, you’re already behind. But they are not the destination. These tools use the basic AI plumbing. Agentic AI is the orchestration that sits on top of that plumbing.”

The nature of agents

Whitehead said agentic AI is technically defined as “a specific arrangement of LLMs, databases, memory, and oversight.” What makes it unique is its ability to run multi-step processes, adapt in real time, collaborate with other agents, and keep moving until goals are met.

“Generative AI is reactive. Agentic AI is proactive,” he repeated. “Generative is about content. Agentic is about outcomes.”

He showed a slide that summed up the shift in Internet use from traditional Web to generative AI and then into agentic in three words: search → answer → action. 

The search era gave you an address you were looking for, the answer era would give you options on a map for getting there, and the action era books a self-driving car to pick you up because the agent knows the appointment in your calendar. 

What this means for media

Whitehead offered a glimpse of the newsroom and consumer experiences agentic AI could enable.

For audiences, agents could become trusted companions: surfacing digests each morning from multiple news sources, reading curated updates aloud via smart speakers, translating or summarising content instantly, booking tickets or buying recommended products directly through a news platform.

“Imagine a news agent that knows you have 10 minutes before your next meeting, and it serves a digest tailored to that window,” Whitehead said. “That’s radical relevance.” It was a term created in a recent workshop in Zurich run by Jodie Hopperton for INMA and Open AI, made up of media executives working in AI. That invitation-only workshop series continues across the world this month.

The impact of internal use of agents at news companies is equally dramatic: research and publishing assistants, automated reformatting of content for multiple platforms, personalised newsletters that adapt based on reader engagement, full-service customer support agents, automated preparation of sales decks and background notes before meetings as soon as they hit the calendar.

“We’re not talking about narrow efficiencies,” Whitehead stressed. “We’re talking about entire new workflows, products, and business systems.”

He then unveiled what he described as the starter list of 12 top opportunities for agentic AI in news media:

• An agentic operating system — a “full formatter” to adapt content across platforms and channels at a fraction of the cost of traditional content management systems.
  

• A content licensing system for AI, enabling publishers to monetise their material in machine-to-machine exchanges.
  

• A personal action assistant to perform all support functions without being asked.

• Full-service customer service management handled by autonomous agents.
  

• Internal technology empowerment — securing, deploying, and scaling AI across organisations.
  

• Agentic research, analysis, and dissemination of insights within newsrooms.
  

• An agentic print production system, adding generative AI tools to round-out the print automation tools that already exist.
  

• Pay-per-view content exchanges for light or occasional readers, with agents equipped with wallets for pre-approved small transactions.

• Subscription content distributors that push personalised feeds across environments.
  

• Next-level ad self-service campaign management with agents handling set-up and optimisation.
  

• Agentic programmatic ad exchanges, buying and selling inventory autonomously.
  

• Business and corporate operations handled by agents, from admin to analytics.
  

Whitehead cautioned the audience not to view these as science fiction: “Don’t see these as futuristic. See them as scenarios to test now.”

He cited as an advanced case the agentic AI adoption at Ippen Media, based in Munich, whose CEO Jan Ippen and CTO Markus Franz have pioneered its adoption in media.

Principles of design

Whitehead urged news companies to design for three principles identified in the INMA/OpenAI workshop: radical relevance, radical proactivity, and seamless knowing.

“Our users don’t want to go from an insight to an action through three clicks,” he said. “They want the agent to close that gap. They want a system that doesn’t just inform them, but acts for them.”

He called this the “agentic AI razor” — a sharper way of ensuring content in all its forms truly meets user needs.

The risks

Whitehead was equally clear about the risks.

• Trust must remain paramount. “We are moving into a world where agents are working on our behalf,” he said. “That is not a world that can function without transparency and oversight.” He offered three models: human in the room (where specialised domain knowledge remains critical), human in the loop (the mode that has become standard oversight of AI outcomes), and Ippen Media’s human on the loop (which requires a much more limited supervisory role of the agents’ work).

• Dependence solely on external platforms was another danger. He pointed to the AI-driven Perplexity browser, already on sale, as a sign of how quickly distribution power could shift away from publishers. “We cannot afford another Google Zero,” he warned.

• Cost is also a factor. Implementation of generative AI is not cheap, but agentic AI itself can reduce costs of implementing news products and content. Whitehead argued the economics will soon change: “We will be able to deploy teams of lower-cost people-like bots. The challenge is cultural — preparing organisations to work alongside them.”

Above all, he warned against professional content getting completely lost in a world of infinite content: “In a world of abundance, relevance is the only scarcity. That is where you must differentiate. This is where we need to focus our tools to sharpen what we produce as humans.”

What news executives must do

Whitehead’s conclusion was framed as an urgent call to action.

“Explore and understand,” he urged. “Lean in now. Create short-term scenarios. Implement test cases and iterate widely. And prepare for the biggest change in your life.”

He emphasised that waiting is not an option. 

“This is not hype,” he said. “This is happening. The organisations that begin experimenting today will be the ones ready to thrive.”

Conclusion

Agentic AI is not just another tool in the newsroom toolkit. It is a general-purpose technology on the scale of electricity or the Internet — only its impact on us will be exponentially faster.

“Agentic AI’s impact on news media will be profound,” Whitehead said. “And it will come sooner than you think.”

YOU MUST BE A MEMBER TO ACCESS THIS CONTENT

Become An INMA Member or Log In

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .