Agentic workflows, intent, and control: what publishers should take from IAB UK’s summit
AI can feel abstract until it starts changing how people discover products, express interest, and ultimately make decisions. At IAB UK’s recent AI Growth Summit, I was impressed at how the speakers managed to avoid hype, fear, or promises that machines will make everything faster and more cost-effective.
Instead, they focused on a more practical question: how can the advertising industry use AI without losing sight of the signals, incentives, and human judgement that sit beneath it?
Recognise that intent starts long before the sale
When it comes to the customer journey, strong examples focused on using AI to capture intent before the point of purchase. Intent shows up in the content people read, the questions they ask, the language they use, and the interests that cluster around a need. Understanding this is crucial because planning improves when teams can spot and interpret demand early enough to influence the eventual purchase.
Nano Interactive, for instance, makes this tangible through their NanoQ Agentic Media Planner, which connects users to its proprietary intelligence layer. The platform uses AI agents to learn campaign objectives, brand values, audience intent, and relevant sector trends, helping marketers build more informed targeting strategies in just a few clicks.
More broadly, the summit highlighted the industry's shift towards agentic advertising, where AI agents do more than analyse data. They are increasingly being used to plan campaigns, optimise media, and surface recommendations autonomously. As media systems become populated by planning agents, buying agents, and optimisation agents, understanding how intent is captured and interpreted becomes even more important.
For publishers, this raises a practical challenge: ensuring their content, audience signals, and contextual data can be understood not only by people, but also by the systems making decisions on their behalf.
Make better decisions, not faster mistakes
A second theme was the danger of treating efficiency as the main goal of AI adoption. Tom Goodwin challenged the industry to stop using new tools just to make old habits cheaper and faster. A poor process doesn’t become a good one simply because it runs at machine speed.
That warning reappeared in more technical form with WPP’s chief AI officer, Dr Daniel Hulme. Publishers often collect more information than they can meaningfully act on, so the challenge isn't simply generating more insight. He encouraged them to focus on the harder task of matching the right algorithm to the right decision, then stress-testing what happens when a system underperforms or exceeds expectations.
For both publishers and advertisers, that creates a practical checklist for AI in advertising. Can a platform explain what signals it uses, whose data it learns from, what outcome it optimises towards, and whether the learning stays with the user? If the answer is vague, the technology may look impressive, but it may deepen dependency rather than improve decision quality.
Prepare for conversational discovery
Search is evolving from a keyword-driven activity into a conversational one. For example, Microsoft’s zero UI framing pointed towards a world where people ask, delegate, and expect useful answers without always moving through the traditional web journey. Websites aren't disappearing anytime soon, but visibility changes when AI becomes the layer between a person and a decision.
Captify’s work on conversational search gave that shift a more commercial shape. People may not know the exact product, brand, or destination they want, but they are already describing needs, problems, goals, and constraints. That creates an earlier stage of discovery, where brands can enter the recommendation journey before preferences begin to solidify.
Meanwhile, Tripadvisor showed why trusted source material becomes more valuable in that environment. Travel is a category where people welcome planning help, but still worry about hallucinations, weak recommendations, and the loss of lived experience. Reviews, local knowledge, specific details, and real human experiences become the raw material AI systems draw on.
For publishers, this shift is about more than visibility in search results. If conversational interfaces become a primary route to discovery, publishers need to think about how their content is structured, surfaced, and attributed when AI systems retrieve information. Clear expertise, original reporting, strong metadata, and trusted audience relationships become strategic assets because they help content stand out as reliable source material for both people and machines.
Embrace agentic advertising, but keep the human layer visible
The most convincing examples of AI weren’t presented as complete replacements for people. Wesley ter Haar, cof-founder and chief AI officer of Monks, showcased how agentic workflows can hold brand context, flag opportunities from cultural and social data, and move from insight to strategy to creative execution more quickly. For instance, with Metropolis, which he described as the first "agentic content supply chain", built for General Motors, specialised agents worked together to identify trends, test creative ideas against audience profiles, and generate variations designed to improve media performance.
The takeaway for publishers is not that humans disappear from the process. Rather, they need repeatable workflows, shared processes, and clear human ownership. As AI agents take on more operational responsibility, the competitive advantage increasingly comes from defining objectives, setting guardrails, and knowing when to challenge the output.
Alongside this, speakers pointed to a broader shift in how work itself is being structured inside organisations. At Publicis Sapient, for example, CEO Nigel Vaz explained his engineering teams are already using multiple AI agents per developer to manage backlogs, break down requirements, and test outputs against original intent. Rather than acting as a single tool, these agents function as a distributed layer of oversight and execution, handling specific steps in a workflow while humans focus on direction and validation. The same model could increasingly apply to adops workflows - from campaign setup and trafficking checks through to performance QA and optimisation loops - where agents continuously monitor whether execution matches the original brief.
Meanwhile, Skin + Me brought the same point into brand production. AI helped with asset creation, repurposing, creative analytics, diversity testing, and reducing dependence on traditional photoshoots. Yet the strongest lesson was about restraint. Real skin results depend on trust, and not every image, avatar, or shortcut is right for the brand.
Finally, Zehra Chatoo of Code For Good Now added an important warning about bias and participation: AI reflects the data, behaviour, and judgement it learns from, meaning adoption gaps can quickly become output gaps. The research she discussed on stage also pointed to women adopting AI at 20% to 22% lower rates, as well as an AI judgement penalty in how AI-assisted work is assessed.
The summit made a strong case for AI that can be inspected, questioned, and improved. The industry doesn’t need another layer of mystery between the user and the decision. Systems should make intent easier to act on, creativity easier to test, discovery easier to understand, and decisions easier to challenge.
For publishers, the opportunity is not simply to adopt AI tools, but to become more intentional about both the signals - and the workflows - they create and control. Trusted content, contextual understanding, audience insight, and transparent data practices are becoming more valuable as both people and AI agents take a greater role in discovery and decision-making.