Let's start with some myth-busting. Which DXP "best practice" is past its sell-by date?
The whole suite versus composable debate. I've sat through ten years of panels on it. Buy one integrated platform, or assemble the best pieces yourself? Underneath, it was always about where the data lives and who owns the interface. Agents are making most of that beside the point.
More and more, the thing using your stack will be an agent with tool access. Not a marketer clicking through screens. And an agent doesn't care whether your architecture is elegant. It cares whether it can find clean, addressable, trustworthy context. A perfectly composable stack with six sources of truth is a nightmare for an agent. So is a big suite with all the data locked behind a UI.
The architecture question hasn't gone away. It's just been demoted. Context quality is the axis that matters now. And I've met very few organisations that have ever measured theirs.
You're freshly certified as an Opal administrator. Straight answer: what can an AI agent be trusted to do on its own today?
It comes down to reversibility. If the agent produces a report rather than an action, and someone would notice within a few weeks if those reports started going wrong, let it cook.
If I were running a marketing team today, my first unsupervised agent would be something boring. A Monday morning audit. It goes through the pages that matter most and flags outdated claims, broken links, missing metadata and copy that's drifted off tone. A product name that changed three months ago. Maybe a page AI overviews have stopped citing. It ranks the lot, drops the list in the team channel, and stops there. Nothing groundbreaking, you can do it today. And it takes over a job nobody was doing properly anyway.
So where does a human still need a hand on the wheel?
Anything that commits. Publishing, spending money, promising something to a customer. And, of course, deciding what we actually want in the first place.
Plenty of CMOs feel like they're losing the "AI race". What would you tell them?
Relax a bit. I don't think anyone is actually ahead. This year I've seen plenty of companies buy a lot of AI with very little to show for it. And a handful of teams visibly pulling away. What the second group has in common is volume of experiments.
Ten small ones on real workflows will get you further than one big autonomous initiative. Each a bit dull, each measurable. The big one takes nine months to reach production, then gets quietly switched off. So treat this as a habit to build, not a purchase to make. Pick one workflow your team already resents doing by hand. Give an agent the report-only version. Put it on the calendar. Then do the next one.
What's the project you've been itching to get your hands on that you’ll finally get to here?
Agents that sit on top of the CRM. For years I've watched the DXP and the CRM live in separate buildings, run by people who meet about twice a year. I've become convinced an agent's decisions are only ever as good as the context it can reach.
The richest customer context usually sits in the CRM. Deal stage, open cases, what they bought, what they complained about last quarter. A content agent that knows an account has three open support tickets will write a very different landing page from one that has no idea. NoA Ignite is one of the few places working on both sides of that wall. So this is where I finally get to build it properly.
Quick one. Most overrated word in MarTech?
Seamless. Same answer for years. Every seamless experience I've seen was held together by a huge amount of invisible stitching. Integration work, data cleaning, someone maintaining the mapping tables at eleven at night. The word describes the result and hides the labour behind it. So whenever a vendor says seamless, ask who's holding the seams together. And what happens the day that person leaves.
If you're right about where this is heading, what does a digital experience look like in three years?
Today it's basically a set of pages with some data behind them. In a few years, that flips. The experience becomes a structured layer of context about the visitor and the offer. The page is just one thing you render from it. Alongside the answer an AI assistant gives about you, the summary a procurement agent reads, or the brief that lands with a sales rep.
A meaningful share of your visitors won't be human. The humans who do show up will already have been briefed by a machine. What doesn't exist yet is an experience that reconfigures itself around intent rather than segments. One that actually knows the visitor's context instead of guessing at a persona.
That's what eventually splits the marketing organisation in two. Specialists who build and maintain the machine, and practitioners who steer it. The tools will turn out to be the easy part. Nobody has designed that team yet.
You say nobody has designed that team yet. Optimizely's Virtual Teammates look like a first attempt. What will you show at DMEXCO that settles it?
They're a genuinely good first attempt at one half of the problem. That half is the AI worker itself. It has its own identity, its own memory and a specific job. And a human approves what it proposes, where that's needed.
The half nobody has designed yet is the human side. Who manages that teammate? Who decides when it gets more autonomy? Who maintains it? And which roles does a team actually need first?
At DMEXCO, I'll show how I'd staff a marketing team with Virtual Teammates. We'll run a live, report-only agent on the stand. And I'll be asking marketers for the one job they'd hand to an agent tomorrow. Because the team gets designed from the jobs people actually want to give away.
Welcome aboard, George!
