If your organization is investing in AI, or fielding pressure to, the conversation has probably shifted from "should we?" to "how do we do this well?"
That second question is a lot harder. And most teams are underprepared for it.
We recently attended a session led by Vitaly Friedman, founder of Smashing Magazine and UX lead at the European Parliament, focused on the state of AI UX in 2026: where the technology is headed, where it's already creating problems, and what it means for the organizations building and adopting AI-powered products.
Agentic AI Is Moving Faster Than Most Organizations Are Ready For
Friedman opened with a projection that reframes a lot of current AI planning: agentic AI (systems that act autonomously on a user's behalf) is expected to surpass traditional development approaches in 2026 and outperform across the board by 2027.
That's not a future-state concern. It's a now problem.
Most organizations are still treating AI as a productivity layer sitting on top of existing workflows. What's actually happening is a structural shift in how digital products need to be designed, governed, and trusted.
Teams that plan for incremental optimization are going to find themselves behind quickly.
AI Doesn't Reduce Work. It Redistributes It.
One of the sharper observations from the session: AI doesn't necessarily make organizations more efficient. It can intensify the workload while creating the appearance of progress.
Output increases. Speed increases. But so does the volume of content that needs to be reviewed, fact-checked, and corrected.
And the errors aren't always obvious. IBM has flagged a growing quality problem: as AI models increasingly consume AI-generated content, which often contains inaccuracies, output quality degrades over time. The signal gets noisier. The cleanup gets harder.
Internally, this creates a real culture problem. When AI-generated output circulates through a team without meaningful review, mistakes get laundered as finished work. Friedman called it plainly: AI slop. It looks done. It isn't.
The organizations doing this well aren't using less AI. They're using it with more structure and intentionality, and they're building in human judgment at the right points.
The Hidden Risk: Approval Fatigue
When AI agents operate with autonomy, they typically ask for human sign-off before taking actions: run this code, send this message, make this change.
At scale, that constant interruption creates a dangerous dynamic: people start approving without actually reviewing. The checkpoints become theater.
Friedman noted that approximately 1 in 6 auto-approved actions carries real risk. In other words, meaningful oversight is already slipping through the cracks in teams that believe they have it.
For digital marketing managers and their stakeholders, this is a governance question as much as a design one. Workflows that look controlled can still fail quietly.
The Trust Problem Nobody Is Designing For
This was the most practically useful framework from the session, and it applies directly to any organization evaluating or deploying AI-powered tools.
There are two ways AI trust breaks down, and they pull in opposite directions.
Automation aversion happens when users refuse to rely on AI even when it genuinely helps them. This usually traces back to bad past experiences, opaque systems, or no sense of control or override.
Overreliance happens when users follow AI recommendations even when they're wrong, because the interface is designed to feel authoritative. Confident outputs. Clean language. Explanations that sound credible even when users have no way to verify them.
The target is what Friedman called calibrated trust: users who know when to lean on AI and when to double-check it.
Here's the uncomfortable part: the same transparency features that build calibrated trust in one context can quietly push users toward overreliance in another. When users can verify what the AI is telling them, transparency helps. When they can't, it becomes persuasion.
That distinction has significant implications for how AI features get designed, and for how organizations communicate AI capabilities to their customers and internal stakeholders.
The Right Model: Acting With Users, Not For Them
The practical design principle that emerged from the session is simple to say and harder to execute: the best AI experiences assist without controlling.
There's a meaningful difference between AI that completes a task on your behalf and AI that helps you complete it yourself.
The first trades short-term speed for long-term risk. The second keeps humans genuinely in the loop, builds trust over time, and reduces the compounding errors that come from unchecked autonomy.
Trust, Friedman argued, doesn't emerge by default. It's built through consistent reliability: understanding the product, using it successfully, achieving real goals, and expanding from there.
This also means interfaces need to go beyond the chatbot pattern.
Open-ended "ask me anything" experiences create what Friedman called articulation barriers: users often don't know how to phrase what they need, which means they don't get what they actually want.
Effective AI UX anticipates that gap, offering structure, guidance, and clear constraints that lower the barrier to productive use.
Questions Worth Bringing to Your Next Internal Conversation
If you're building a case for AI investment or trying to make sense of what you've already deployed, these are worth asking:
Which parts of your digital experience actually benefit from AI, and which are better left as conventional UX?
Where does human oversight exist in your AI workflows, and is it real, or has it become a rubber stamp?
How is your organization designing for the moment when AI gets it wrong?
Are you building toward calibrated trust with your users, or are you accidentally engineering overreliance?
These aren't theoretical debates. They shape how teams work, how expertise develops, and how organizations create value over time.
We're paying close attention to where this is headed, and we think more companies should be having these conversations now, not later.
If you're looking to get your site better positioned for GEO and AI-powered search, let’s talk.
