Who would you hire if you were picking your marketing team from scratch?
Team members that carefully strategize their work ahead of time, proactively collaborate and know when to look for external help to achieve the objective, with the right level of autonomy?
If this sounds good, then you will love AI agents. And they can get to work in 2025.
Unlike traditional automation workflows that follow a predefined linear decision tree, you give agents a complex task, and they define the sequence of actions to reach the goal, call third-party systems and delegate sub-tasks to other agents. The concept has been around for over 18 months already.
Agents are autonomous and thoughtful, incredibly fast and knowledgeable. They won’t replace star performers, but they will supercharge your team’s delivery.
A revolution under way
AI agents are already everywhere: Salesforce has Agentforce for CRM teams; Apple launched personal agentic assistant Apple Intelligence; Anthropic released computer use, an agent that automates mundane tasks through your browser; and Google has a similar product with Project Mariner.
Ask an AI agent what to wear to Google Cloud Next conference, for example. With function calling – the ability to use a third-party API with natural language instructions – it understands it’s a Las Vegas conference in April. It then calls a weather forecast API with location and dates to know the likely temperature and humidity.
However, there’s a world between cleverly prompting an LLM and designing true AI agents. You need upfront planning and autonomy, which, until now, had not been totally cracked, making most AI agents nothing more than natural-language-based workflows.Things are changing fast, though.
Marketing inefficiency and AI agents, a match made in heaven
Going back to the definition of AI agents – an autonomous technology system that plans and executes a multistep workflow with limited human supervision in response to natural-language-based instructions – marketing use cases easily come to mind.
Create better briefs by dynamically populating the right fields with the right information. Automatically detect and fix anomalies in campaign tracking naming conventions. Fully automate website QA processes by asking an agent to test specific customer journeys, taking screenshots of frontend issues before go-live.
And what if you could improve audience planning? In the addressable media age, campaigns need to be personalized to specific audiences and mapped with key target groups. However, the process for defining and mapping a set of personas to actionable, large but specific-enough audience segments that are exposed to personalized creatives across media platforms is inefficient.
Most campaign briefs only contain high-level audience definitions, which creative and media agencies struggle to augment with their methodologies and tool sets.
In an agentic world, a master AI agent is prompted with details of a campaign objective, budget and geographic scope in natural language, then instructed to come up with a cohesive audience strategy that agencies use as a source of truth. It establishes a game plan and decides whether to collaborate with other AI agents.
For instance, a research AI agent sifts through company documents to extract target audience insights. A second agent scans consumer reviews and socials from the brand’s community for information to complement market research. A third agent translates audience characteristics summarized from the first two agents’ output into structured queries against the company’s customer data platform, identifying which first-party audiences to target and use as seed audiences for lookalike modeling.
The master agent wraps up each agent’s output, writes a structured report on target audience preferences, brand perception and how to best address them across media inventories to maximize performance. It’s powerful and efficient.
Automation + augmentation
Like human teams, AI agents deliver the most value when specialists work with generalists under supervision of a manager. Like humans, they need access to clean, abundant data. And while you would expect reliability, this is where caution is currently advised.
Because agents are not programmed with total control over what they can and cannot do and because we don’t fully understand how the large language models underpinning them work, it’s impossible to guarantee the same request will generate the same results, or that agents won’t go off-track.
Add in that agentic workflows can be compute-intensive and that AI developers needed to configure them are scarce, and you understand why agents didn’t already revolutionize marketing.
Make no mistake, however, agents are much more reliable than a year ago, and marketers are more accustomed to generative AI. In our own pilots for social, we’ve seen creative teams delivering 40% reduction in time from client brief to boosting content, with triple the creator and brand accounts covered in social audits, thanks to our AI agents.
Cost pressure on efficiencies is not going away, and marketing ecosystems – with multiple workflows, partners, structured and unstructured data – are ideally suited for AI agents.
By pairing automation with augmentation, there is little doubt AI agents will become ubiquitous in marketing organizations. While CMOs should maintain a healthy dose of skepticism around the hype, they should lose no time to ready their organizations for this new, powerful capability by testing at small scale with the right partners.
Like many aspects of AI, the technology is rapidly progressing – and fortune favors the fast.
“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Follow The Brandtech Group and AdExchanger on LinkedIn.
For more articles featuring Hugo Loriot, click here.
