Agentic AI: How agencies are successfully using the phenomenon
Enough chatter - how is adland actually using agents in practice?
Agentic AI: How agencies are successfully using the phenomenon
Enough chatter - how is adland actually using agents in practice?
Agencies are mostly using agentic AI to take the boring stuff off human plates. The early wins are not flashy creative breakthroughs, but quieter gains - removing repetitive, data-heavy and low-value tasks so people can spend more time on judgement, strategy and problem-solving.
Whenever I hear the phrase “AI agent” or “agentic AI” I can’t help but picture little people in tuxedos, armed with handguns, running around inside … wherever it is AI actually exists (a complicated technological infrastructure, nevertheless). But putting my admittedly childish imagination aside, there are real questions here: what exactly are these invisible operatives? How do they differ from the generative AI chatbots most of the world has adopted? And how is the advertising world using them right now?
The LLMs of this world, along with other generative AI platforms, have already transformed how agencies create campaigns, right through from idea generation and planning to media buying and production. But despite their impact, these tools are largely reactive: they respond to prompts, operate in isolation from other systems and have little memory of past context. This is where AI agents come in.
By combining LLMs with memory, planning and system integration, agents can be embedded into real agency workflows. They can work toward defined goals, break tasks into steps and take action across tools and platforms with minimal human input. This shift turns AI from a passive assistant into something more proactive - capable of monitoring workflows, triggering actions and supporting complex, multi-step processes that go far beyond the reach of first-generation generative AI.
But, of course, they are not a “magical panacea”, as Daniel Hulme, chief AI officer at WPP, recently told me. Knowing why, how and when to use agents is a whole different ball game. For the industry, the new rule of engagement is “strategic orchestration,” as Lex Bradshaw-Zanger puts it - Chief Marketing and Digital Officer of SAPMENA Zone for L’Oréal Groupe. For him, it is about:
“Knowing when to deploy AI, how to combine it with human insight, and maintaining control over your data and brand integrity while scaling at unprecedented levels.”
Across global organisations beyond that of advertising, McKinsey research suggests the real promise of agentic AI lies not in the agents themselves, but in how organisations redesign work around them. Rather than layering agents onto existing processes, the shift requires more deliberate design: deciding which tasks benefit from autonomy, where human judgement must stay in control, and how different systems and tools should connect.
Crucially, the biggest challenge is not technical capability but trust, governance and adoption - getting people comfortable with AI that does not just respond, but acts. How this balance is struck in practice is now becoming the defining question for agencies experimenting with agentic AI.
So how are agencies beginning to use agentic AI in practice, and what have they learned about where it adds value, where it struggles and how best to integrate it into day-to-day operations?
Freeing up humans
Dominic Palmer
Director
Rock Kitchen Harris
At RKH we’re using agentic AI to clear the jobs clients don’t value and teams don’t enjoy. Tasks like pulling financial data from different sources, spotting anomalies or assembling reports are ideal.
Agents move through those steps faster and more accurately than a human, which frees up time for work that requires judgement, innovation and lateral thinking.
Fergal O’Connor
CEO and Founder
Buymedia
We see the strongest value where work is rules-based, repetitive or data-heavy, and where people are acting as glue between systems. Planning agents that pull audience and performance data, check constraints and draft media plans are one example. Internal agents that help with coding, summarising meetings or tracking tasks and dependencies are another.
In these cases the AI takes care of the grunt work and frees people to focus on judgement, relationships and creative problem solving.
Carl Ebanks
Digital Director
Cogent
At present, this is something we are investigating and testing but have not fully implemented across the agency yet. An example would be to take digital ad performance data, export that, then use ChatGPT's Deep Research tool to analyse the data and produce an insights document from it, we are then testing AI presentation tools (Genspark, Gamma etc.) to take that report and design an easy-to-digest summary report that can be exported into PowerPoint.
Nick Tong
Managing Partner, Media & Data
Mediaplus UK
Agentic AI manages multi-step tasks like researching, planning, creating and optimising campaigns without constant human input. At Mediaplus UK, we use it for behavioural audience research, media planning, automated reporting and creative variations. Our most advanced applications are brief decomposition, audience research, automated strategic framework application and advanced pattern recognition.
When a client brief arrives, our agentic workflow analyses it, identifies explicit asks and implicit needs, flags tactical traps and surfaces strategic opportunities. It then connects directly to tools like Global Web Index to conduct audience research, pulling insights and generating visualisations in under an hour, work that previously required multiple teams and days of coordination.
We've found it most valuable for enhancing task quality, not just automating repetition. Analysing large datasets to identify content gaps or connecting disparate insight sources happens faster, but humans still decide which topics fit the brand strategy and creative direction. The real gain is creating headspace for deeper strategic thinking and testing creative ideas immediately.
Jonny Goodall
Chief Design Officer
Bernadette
Like many agencies, we initially used AI to speed up thinking and output across research, production and synthesis. Building on that, we’re now applying agentic AI in a more intentional way - creating small, goal-driven systems that run defined workflows, make decisions within clear boundaries, and return control to humans at the right moments. In practice, that means agentic workflows supporting production and synthetic research, including dynamic models of personas and scenarios.
Ant Kenny
Head of Digital
Boutique
We’re using agentic AI to streamline work that would otherwise absorb disproportionate time. Ultimately, as a business we sell time and thinking and with budgets under increasing pressure, our focus has been on using AI to remove low-value effort so more of the budget goes into work that genuinely moves the needle for our clients.
Some of the areas we’re using it more commonly are using AI for research and insight acceleration. We’re using agentic workflows to accelerate early-stage research and insight generation, particularly where the work is structured and repeatable. Examples include automating topical and intent mapping as part of keyword and content research, which used to take several hours, supporting competitor and category analysis by quickly summarising what others are doing, where they’re showing up, and what themes dominate the conversation, and quickly summarising large sets of performance data or coverage reports to spot patterns and trends without wading through spreadsheets.
This has significantly reduced time spent on page-one research and freed teams to focus on interpretation rather than extraction. A bonus too is that it very often helps spot patterns that a human cannot, so our insight has improved considerably across many areas that has enhanced strategy and performance for our clients.
Using AI to make audits and planning more efficient is another key area. Audits and planning are where AI has worked hardest for us, helping us pull together data, spot gaps, prioritise actions and get to a strong first set of recommendations more quickly. This doesn’t replace expertise, but it does bring consistency and discipline to the process. Everyone starts from the same baseline, which makes audits quicker to produce and easier to sense-check, compare and improve.
Using AI to support creative and comms work is deliberately constrained and our use of AI in creative and communications is about support, not substitution. Typical use cases include idea generation for social, PR angles and creative routes, structured draft writing for audits, QBR commentary and pitch materials, and reworking and refining copy in stages to improve tone, clarity and readability. For us, the value is a clear framework that helps get to a strong first draft quickly, with final judgement and sign-off always handled by the team.
Beyond delivery work, we’re also using AI to improve how the agency runs. This includes analysing briefs, tenders and RFPs to extract requirements, risks and resourcing needs, supporting team allocation by highlighting skill requirements and workload patterns, feeding qualitative pitch feedback into structured analysis to spot patterns and improve future performance, and sense-checking documents for errors, inconsistencies and clarity before they go out the door. These uses are less visible externally, but they’ve had a meaningful impact on efficiency and quality internally.
Ollie Sloan
Head of Strategy
Arke Agency
At Arke, we use agentic AI as an internal efficiency engine rather than a front-facing creative layer. Each agent is assigned a narrow, explicit task designed to remove friction from day-to-day operations: research, category and competitor mapping, audience signal analysis, media performance monitoring, content variant analysis and suggestions, and optimisation diagnostics. These agents are not generalists. They are purpose-built operators that have learnt from clear inputs, work towards defined outputs, and operate under hard constraints.
The real value shows up in speed and efficiency.
Agents handle sequential, repeatable, and data-heavy work that previously consumed senior attention and exhausted capacity.
Strategy, creative, and client teams receive cleaner inputs, clearer diagnostics, and more actionable recommendations. Human effort shifts from assembling materials to reviewing and making judgments.
Thomas Walters
Chief Innovation Officer and Co-Founder
Billion Dollar Boy Group
Artificial intelligence has been transforming the advertising industry for years, but a new evolution in AI is turning heads – and changing workflows.
Unlike traditional models that wait for a prompt, agentic AI is autonomous, goal-driven, and capable of taking meaningful action on its own. It can research, plan, create, and optimise campaigns with minimal human input, making decisions and learning as it goes.
At Billion Dollar Boy Group, we’re already using it to transform how we think and react in social-first advertising.
Agentic AI handles the heavy lifting. This frees our experts to focus on what machines can't: building relationships and sparking creative brilliance.
Take something as simple as analysing social media comments. Rather than spending hours sifting through spreadsheets, assessing sentiment, and categorising mentions by brand, product, or creator, our AI agent does it in minutes. It reads tone, understands emojis, and identifies where fans sit in the sales funnel.
Being fluid with workflows
This is also where things can go wrong. Treat agentic AI like a magic plug-in for rigid processes and it quickly becomes more trouble than it is worth. Agencies finding real value are redesigning workflows around flexibility, letting agents join dots across systems, while being clear about where classic automation or human intervention still makes more sense.
Fergal O’Connor
CEO and Founder
Buymedia
At Buymedia we have moved from simple prompt-based experiments to using agentic AI as a coordinated system. At the top level we use an orchestrator agent that understands the overall goal, then routes work to specialist sub-agents for planning, purchasing, managing and reporting. Each of these sub-agents can call lower-level skills and tools, such as APIs, databases or search, so the AI can actually carry out tasks rather than only suggest ideas.
Nick Tong
Managing Partner, Media & Data
Mediaplus UK
Our biggest challenges have been integrating agentic AI with existing tools and building team trust in outputs. The best approach has been setting clear guardrails, connecting workflows to existing platforms and maintaining human review at every stage. We're currently running regular hackathons to stress-test how AI can fully integrate into pitch and planning processes, with early results showing significant efficiency gains that translate into more creative exploration and client growth, not reduced costs.
Sarah Treliving
Chief Digital, Data and Technology Officer
Goodstuff Communications
Agentic working starts with the automation of a process or conversational access to output of a tool but exponential value comes from the ability to join up processes to deliver more ‘point’ or ‘end to end’ solutions that create cross-function capabilities. This connectedness or ‘multi-agent reasoning’ requires AI that references multiple systems, which can be a challenge depending on the data and technology decisions of the past.
In general, and in the ideal, the more proprietary the source system, the greater the flexibility to be able to extract, pivot and supply data for AI. As with previous technology evolutions, it is ‘meta’ technology that allows AI-powered systems and sources to talk to each other. Happily, these are coming to market, at lower costs than third party legacy systems. Now, more than ever it's key that data and AI strategy is not solely reliant upon locked into mega-silos from internet/AI giants.
Dominic Palmer
Director
Rock Kitchen Harris
The mistake is assuming agentic AI is a panacea, when in reality it isn’t suited to everything; if a task is essentially ‘if this, then that’ then classic automation still does it better.
Agentic systems are designed for goal-driven problem solving, not perfect execution of a rigid script, so forcing them into fixed-rule workflows only creates friction.
Thomas Walters
Chief Innovation Officer and Co-Founder
Billion Dollar Boy Group
Our agents vet deliverables against briefs and suggest creative tweaks. We also integrate AI into Companion, our proprietary software, to find the perfect creator partners. Every agent learns and improves over time, but always with human oversight.
The benefits are clear. It brings efficiency, uncovering insights that would otherwise be buried in mountains of data. It improves decision-making, giving us confidence in our recommendations to clients. And, by handling the repetitive tasks, it lets our teams create work that truly resonates.
Ollie Sloan
Head of Strategy
Arke Agency
The most valuable capability is not prompting but workflow design: defining tasks precisely, setting guardrails, establishing stop points, and diagnosing failure modes.
Agentic AI is treated like a junior executive with extreme efficiency, endurance and speed.
Human touch
Despite all the talk of autonomy, no one here is handing over the keys. As agents take on more responsibility, agencies are doubling down on human oversight, building in checkpoints, reviews and accountability to make sure speed does not come at the expense of trust, accuracy or judgement.
Carl Ebanks
Digital Director
Cogent
At each step we are using our human expertise in paid media to ensure the report is data accurate and that the insights generated are what we would recommend. We will obviously modify the final output and ensure everything is working factually. We feel this move to semi-automated reporting allows us to spend more time on maintaining and optimising campaigns. We are not yet at the stage where these do the work for us completely.
Thomas Walters
Chief Innovation Officer and Co-Founder
Billion Dollar Boy Group
Agentic AI isn't flawless. Creative industries are full of nuance, and AI still struggles with complex emotion and cultural subtleties.
Technical limitations also crop up. All agents are reliant on the server and we have had times where this is ‘at capacity’ and so processes return to manual or are put on hold.
These challenges underscore an important lesson: while agentic AI can do a lot, it can’t be relied upon solely. Ultimately, AI is a tool, and like any tool, its value depends on the skill, judgement, and oversight of the people using it.
The key is knowing when to let the system act and when to step in.
Dominic Palmer
Director
Rock Kitchen Harris
The other watch-out is complexity. Once a workflow has branching logic or multiple dependencies, agents can generate loops, hallucinated steps or over-engineered code that technically runs but isn’t fit for use. That’s why we keep technical oversight in place. You need people who understand how the agent is chaining tasks, writing functions and making decisions. Use agents where flexibility is an advantage, avoid them where determinism is better, and keep humans in control of the architecture and judgement so the technology helps rather than hinders.
Ollie Sloan
Head of Strategy
Arke Agency
Where we’ve found agentic AI performing best, is in environments with clear objectives, measurable signals, and feedback loops. Performance optimisation, research, and operational planning often result in the most effective performance. Where we've come across issues of agentic AI struggling is in ambiguity, subjective decisions, and contexts where success cannot be cleanly defined. Without constraints, we’ve found that agents often optimise the wrong variable.
For Arke and our clients, trust has been the central integration challenge. Our model enforces human review at two explicit points. The first occurs during individual agent analysis, diagnosis, and recommendation, where humans validate logic, assumptions, and framing. The second is a final human review before any output influences creative direction, client recommendations, or strategic decisions.
Agents never self-approve.
The core rule for Arke is that accountability has and will continue to remain human at Arke and to be disciplined in deployment when utilising AI in this way. Specific tasks, layered review, transparent reasoning, and enforced human checkpoints. Used this way, agentic AI compounds efficiency without eroding judgment.
Ant Kenny
Head of Digital
Boutique
Most of our use to date has been practical rather than fully autonomous.
We’re not handing over decision-making, but we are using agent-style workflows to run defined tasks consistently and at speed, with humans staying firmly in control.
We’ve found AI adds most value when it’s applied with clear boundaries. It’s strong at repeatable tasks and has reduced time spent on low-value research and admin. It’s less reliable where judgement or nuance is required, or where data is weak or highly specialised, so in those cases it supports thinking rather than drives decisions.
What matters most for us is how AI fits into day-to-day work. There’s always a clear owner, everything is sense-checked, and we’re clear on when to fall back to human judgement as AI is there to support our existing processes and not replace them.
Fergal O’Connor
CEO and Founder
Buymedia
The real challenges are around context and integration, not the model itself. You have to define exactly what data each agent sees, what a good outcome looks like for its role, and how context is passed from one agent to the next so nothing is lost. We keep a human in the loop at key points. A separate evaluator AI can score and critique outputs, but people still make the final call.
AI innovation
Beyond efficiency, agentic AI is also opening up new creative territory. Not just in how agencies work, but in how brands show up, allowing AI to act on a brand’s behalf inside services and experiences, not just speak for it in communications.
Ollie Sloan
Head of Strategy
Arke Agency
A critical evolution in our setup has been the introduction of an overarching supervisory AI. This system does not generate work, but it analyses the outputs of each tasked agent, looking for gaps, inconsistencies, weak assumptions, and optimisation opportunities. In effect, it acts as that first point of review, highlighting anomalies, and ranking confidence levels before any human time is spent.
Thomas Walters
Chief Innovation Officer and Co-Founder
Billion Dollar Boy Group
Winning with AI is about culture, not just code. We build trust by showing our teams tangible wins. Every agent starts in a controlled "sandbox" environment with clear success metrics. We then test it repeatedly on the same data to refine its performance and establish guardrails.
Once a tool is ready, we train staff in its capabilities and appoint AI champions in every team to maintain a feedback loop. This ensures we deploy technology thoughtfully.
Integration is not just about technology; it’s about culture. When teams see AI as a partner that amplifies their work, the fear vanishes.
We’re only beginning to scratch the surface. Instantaneous reporting, rapid influencer campaigns, and intelligent influencer vetting are just the start. The next frontier will see AI agents collaborating across teams, learning from each other, and helping agencies respond faster and more intelligently than ever before.
Jonny Goodall
Chief Design Officer
Bernadette
Where it becomes genuinely exciting is how agentic unlocks new ways for brands to exist inside digital services, not just in communications. Traditionally, a lot of investment goes into defining brand characters and personality, only for that richness to disappear once customers enter a service journey.
Now with Agentic AI, the AI acts on the brands behalf, not just as the user's.
So the brand now must speak through what it does, not just what it says.
In agentic services timing, intent, and delivery become the new tone of voice - and a key brand differentiator.
To answer this, here at Bernadette we developed a methodology for our clients focusing on this called 'Tone of Action' to identify those points of agentic benefit and brand opportunity to stand out in the market. The result is ownable “branded” agents and service moments that simplify digital experiences, differentiate from competitors, and allow brand DNA to live consistently across the entire ecosystem and customer experience, in turn evolving brand-defining characters into AI-powered personalities, expressed through dynamic conversation and behaviour rather than messaging and mascots alone.
Built with Shorthand