Andy Cohen is the founding architect of Sitecore XM Cloud. He is VP/Chief Architect at Altudo, and a CMS Critic contributor.
When people talk about personalization, they often mean “the right content to the right person at the right time.”
That idea is now past tense.
I have watched this shift play out up close, from Sitecore’s early rules-based personalization, through Contentstack’s use of edge-based personalization in a composable architecture, to Optimizely’s launch of Opal, a system of marketing AI agents.
Personalization no longer stops at swapping content on a page. It now reaches into the interaction itself, and soon, into the relationships agents form with one another.
That shift, from “what” to “how,” is where the major CMS players are already moving, and you can see it in the different bets they're making:
None of these approaches feels like a gimmick.
They are different answers to the same problem: personalization is no longer about what content you are shown, it is about how the interaction unfolds. Whether that happens through hyper-personalization at the edge, through orchestrated marketing agents, or via copilots and execution partners, the shift is real and already in motion.
Most software behaves like math: 2 + 2 always equals 4. Generative AI does not. Ask it to draft a thank-you note and you will get three different answers, all plausible, none identical. That variability is what we call nondeterminism.
It is the very thing that makes AI feel creative, but also the reason it needs personalization to stay useful. As Pariveda Solutions points out, nondeterminism is what makes AI flexible and creative, but it also introduces governance challenges that remain unresolved. In my own work with XM Cloud, I have seen how unpredictable outcomes can be a feature if the system adapts well – and a liability if it does not.
The next frontier is agents adapting not just to us but to each other. Nondeterminism multiplies when two autonomous systems interact. They need ways to converge on workable outcomes.
Sometimes that looks like adjusting communication style, where one agent is structured and terse, another free-form and verbose. Sometimes, it means learning to accept differences in task style, such as fast approximation versus cautious step-by-step reasoning. Other times it's about role negotiation: who leads, who validates, who escalates.
And sometimes it is memory. Agents do not just adapt in the moment; they remember. That memory can look like a grudge. A search agent that hallucinated last time will get fact-checked the next. A procurement agent that botched a deal may face stricter oversight. Even first impressions stick. If an agent’s very first hand-off is sloppy, the other agent may never fully trust it, no matter how good the later work becomes.
Humans behave the same way. In corporate life, one bad kickoff meeting can color a partnership for months. Agents will carry those patterns too. Deepset calls this the “spectrum of determinism to nondeterminism.”
I call it relationship management.
Picture a grocery agent told to buy apples if they are $0.99 a pound. A deterministic system comes back empty because they are $1.09. An adaptive one checks with you, “They are a little higher today. Should I buy anyway?” That is useful, until the store’s own agent learns to nudge yours into paying even more.
Or ride-hailing. You set a $20 target. The price surges to $23. Do you want your agent to decline, wait, or decide on its own? What happens when the ride-share’s agent starts upselling yours into priority pickup at $28?
Or travel. You say, “Book me a flight under $500.” The system finds none. A flexible agent returns with, “There is one at $525 with better times. Should I grab it?” Helpful, but it shows how quickly rules can bend.
In the enterprise, this gets thornier. Procurement policies might cap software licenses at $50 per seat. A deterministic agent halts the deal. A personalized one negotiates to $55 and closes it. That looks like a win until you realize the vendor’s agent deliberately anchored higher, knowing yours would flex upward.
Healthcare raises the stakes even more. You ask for an appointment with your doctor next week. A rigid agent says none available. A personalized one offers a nurse practitioner tomorrow or a specialist in two weeks. That looks like a better experience, but it could also push you toward higher-margin services instead of the care you really wanted.
Salesforce has written about this exact problem of when to script agents tightly and when to let them roam. I see it every day in delivery projects. Too rigid and you frustrate users. Too loose and you open the door to manipulation.
Now push it further. Imagine your agent not only executes tasks but carries your personality footprint into every interaction. If you are cost-sensitive, it pushes harder. If you are time-poor, it pays more for speed. If you are detail-driven, it double-checks everything.
What if you could swap personas entirely?
Your agent would not only personalize to you. It would personalize as you.
But there is risk in this. If the other agent knows who you are or which persona you are using, it can adapt back. That can be powerful when aligned with your goals, but it can also be exploited. Imagine a vendor agent that recognizes executive mode and steers you toward premium packages.
This is not hypothetical. Optimizely’s Opal agents already act like different team members depending on context. Contentstack’s Brand Kit lets you define voice profiles and store guidelines so that every AI-generated artifact reflects your identity. Sitecore’s Stream copilots are designed to guide content creators in real time, aligning their work with brand and role expectations.
These are early signals of a future where agents carry pieces of us, including quirks, preferences, and even biases, into every exchange.
This is why nondeterminism is not a flaw but the feature. Deterministic systems did not need personalization. There was only one output. Nondeterministic systems can go in many directions, and personalization is what prevents chaos.
The best metaphor I know is jazz. Musicians improvising do not know exactly where the others will go. They adapt, they respond, they personalize their playing to the group. The music works not because it is scripted but because the players trust each other enough to bend and flex.
McKinsey has argued that this adaptability is the defining advantage of agentic AI. From what I have seen in publishing systems, SaaS cost optimization, and large-scale CMS migrations, I agree. The resilience comes not from rigid plans but from the ability to adapt in context.
Leaders should focus on their readiness in key areas. For example, trust frameworks must extend beyond human-agent interactions into agent-to-agent ecosystems. Standards cannot stay as fixed APIs; they need flexibility built in. Identity management has to protect personality profiles and alter egos from misuse. And agents must be able to remember not just what they did for us, but how they worked with others.
Personalization started with content. It evolved into experiences. Now, it is moving into interactions – first with humans and then with other agents.
If we get this right, nondeterminism will not be something to suppress. It will be the source of resilience, adaptability, and creativity. The companies that thrive will not just personalize their websites. They will personalize their agents. Those agents will adapt to us, and to each other, in ways that reflect who we are or even who we choose to be.
And like humans, they will remember. First impressions will matter. Grudges may form. And relationships between agents will start to look a lot like relationships between people.
October 14-15, 2025 – Amsterdam, Netherlands
November 4, 2025 – New York, NY
November 6, 2025 – Los Angeles, CA
Join us at JoyConf 2025, where the future of content meets the people shaping it. Built for developers, marketers, and digital innovators, this free conference goes way beyond product updates or yet another strategy playbook. It’s where you’ll get inspired, get real, and get connected – to ideas, to the community, and to what’s possible when you build joyfully. Whether you're building lightning-fast frontends or launching cross-channel campaigns, this is where content becomes connection and collaboration becomes second nature. Come for the talks. Stay for the community. Leave with joy. Save your spot.
November 3-5, 2025 – Orlando, FL
At Sitecore Symposium, you’ll learn how to meet your customers where they are – across channels, moments, and touchpoints that didn’t exist a year ago. You’ll hear how the boldest brands in the world are transforming content into outcomes and creating experiences that are fast, relevant, and human. Join 1,500 marketers, digital leaders, and technologists in Orlando to see what’s possible when personalization scales, AI accelerates creativity, and trust becomes your competitive edge. This is the Third Wave of the Internet. And this is your moment to lead in it. Get your tickets today.
January 13-14, 2026 – St. Petersburg, FL
Meet industry leaders at our fourth annual CMS Kickoff – the industry's premier global CMS event. Similar to a traditional kickoff, we reflect on recent trends and share stories from the frontlines. Additionally, we will delve into the current happenings and shed light on the future. Prepare for an unparalleled in-person CMS conference experience that will equip you to move things forward. This is an exclusive event – space is limited, so secure your tickets today.