TL;DR: Most agentic AI strategies fail not because the technology is bad, but because teams build agents without coordination, verified data, or direct distribution, optimizing for activity over impact. Chris Brownlee, SVP, Product at Yext, explains why brands that win in the AI era won't be the ones with the most agents — they'll be the ones whose agents work together from a single source of verified truth.
The phrase "AI agents" is dominating headlines — to the point where the words have started to mean everything and nothing.
Every software vendor has them (or claims to). Every company is talking about them. And a lot of companies have spent the past year deploying them, only to find that the results don't quite match their expectations.
But that doesn't mean the technology is broken or that AI agents don't have the potential to change our work for the better. It's just that there's a meaningful difference between "having AI agents" and having an agentic strategy that yields results.
Having a handful of task agents firing off review responses, updating listings, or scheduling social posts can look like progress. But if those agents aren't coordinated, aren't working from accurate data, and aren't grounded in goals that extend beyond their individual tasks, you haven't built an agentic strategy. You've built a more automated version of what you already had.
So, what should brands be doing instead? We asked Chris Brownlee, SVP, Product at Yext, to break down why "more automation" too often fails — and what you can do to avoid these pitfalls.
Lauryn Chamberlain, Senior Editorial Strategy Manager: Chris, every company says they have "AI agents" now. What's the version of that conversation you're tired of having — and what do you wish more people were asking about?
Chris Brownlee, SVP, Product: The term can sound technically obscure, but once you picture an AI agent like another employee who can do jobs for you, it starts to make more sense. The problem is that any product today can apply an LLM to a task and call it an agent — just like you can technically hire anyone off the street for a job.
But we don't just hire anyone, right? We look for experience, knowledge, domain expertise. We look for someone who can make good decisions based on the data and facts in front of them. Basically, automation is like an entry-level employee: it can only act on the task directly in front of it. A true agent is more like a Digital Marketing Director, who can handle a much higher degree of ambiguity.
Putting an LLM in a product is an off-the-shelf capability now. Any software can do it. The real question is: what context does the agent have, and what are the bounds it's given to work within?
Those bounds matter more than most people realize.
Here's a real example of how this goes sideways. Say an agent identifies that an ice cream shop could reach a broader audience by changing its primary category on Google. With very little context, it might change the category to "Coffee Shop" — because the shop does sell coffee, and more people search for coffee. That seems logical on the surface. But now that ice cream shop is competing with hundreds of coffee shops nearby. And at a tactical level, changing a primary category requires Google to re-approve the profile, which can take days or weeks. That's a disaster.
The less data and context an agent has to work from, the more likely it is to make the wrong move. Is your agent following generic best practices, or does it actually know how you compare to your direct competitors and what specific actions will move your numbers? That's the question worth asking.
You said "data" and "context" a couple of times just now, so that's a good segue: You've talked publicly about structured data as the foundation for AI visibility. But most marketers approach AI agents as a workflow problem, not a data problem. Where does that disconnect come from — and why does it matter?
AI agents can help you build a full, rich data set, and that data is what drives AI visibility. At Yext, we structure everything about a brand into a Knowledge Graph. That data makes you show up consistently across the web, which builds trust with AI search engines. When an AI search engine sees consistent, accurate information about your brand, it's more confident sharing it with users.
The challenge is that keeping that data complete and current is genuinely hard. Managing it manually across thousands of locations, on a daily basis, is nearly impossible. That's why we've built agents whose job is to go out and find information that makes the Knowledge Graph richer: by scanning the web, combining existing data in new ways, or drawing on customer interactions to fill in the picture.
The more an AI search engine knows about your brand, the more likely it is to match you with a customer who is actively ready to buy. High-signal matches connect your brand to the highest-intent customers. That opportunity starts with the data, not the workflow.
When you're evaluating whether an agentic AI strategy is actually sound, what questions would you ask?
The point of an agent is to save you time and increase the value of your brand, just as you hire an employee. You bring in entry-level employees for low-consequence, low-ambiguity work. The same logic applies to agents: highly repeatable, well-defined tasks are where they shine, and we've solved a lot of those for our customers.
But we're now in a phase of building what I'd call middle management agents — agents with access to deeper, broader data sets, the ability to cross-reference information, and the awareness to understand what Agent 1 is doing and how that should change what Agent 2 does next.
The real question to ask of any agentic strategy is whether agents are operating independently or in coordination. Independent agents, no matter how many you stack up, are capped at the sum of their individual tasks. Coordinated agents can do something more interesting.
Walk me through what it actually looks like when agents aren't coordinated. What goes wrong, and why is it hard to see until it's already causing damage?
Let me use a positive example this time to illustrate what becomes possible when coordination actually works.
Say that same ice cream shop has two agents: one that responds to reviews, and one that creates social posts. Each one hums along doing its individual job. But here's what's possible if they can share information the way two colleagues might.
The review-response agent notices that customers keep raving about the flavor of the month. It passes that along to the social agent: "Hey, people are really loving this right now." The social agent takes that insight and builds a campaign around it: posts, timing, and angle all informed by what real customers are saying in real time.
If those two agents can't share information, that opportunity just disappears. The more you can orchestrate agents to work together — handling ambiguous tasks in ways that align with the overall goals of the brand — the more powerful the whole system becomes.
So, where is Yext actually headed with this? Not the roadmap version — what does winning really look like?
It's that orchestration layer. We're working on more individual contributor agents, but also on the middle management layer that can coordinate and adjust strategy based on signals from one part of the brand that might affect another.
A brand's end-to-end marketing needs to work together. To give a customer a great experience across their entire journey, everything has to be coordinated; it has to understand goals and intent. The better that experience, the more likely that customer becomes a paying one.
Any final thoughts for brands evaluating agentic AI right now?
Agents aren't a checkbox feature. Dig into the data and the bounds agents can operate in before you commit. Make sure you're not being sold entry-level employees for a Director-level job.
