The Agentic era of UX
Marco van Hurne
I build AI companies | Data Science Strategist @ Beyond the Cloud | Data Strategy Certified | AI Compliance Officer Certified
If you remember when “web experience” and “mobile experience” were the biggest things in tech, you probably have a receding hairline and a hearing device.
But the good news is that with age comes wisdom. You knew exactly what to expect: pages that looked great until you actually needed them, at which point they whispered, Surprise, here’s your 404 journey!
But then someone asked, “What’s next?”
And just like that, we stumbled head-first into the glorious age of Agentive Experience, where machines assist you and they shadow you like an overly committed life coach who might also be spying on you.
Welcome to the age of ….
AGENTIC AI
… and how to heck design a UX for them agents
… or not, and just offer an API. Read: We need to develop a vision for Machine Customers
Before we start!
If you like this topic and you want to support me:
In the era of Agentic UX
So, I drank the Kool-Aid of agentive technology, some time ago. I spent about a whole year playing with platforms like Crew AI, and Autogen from Microsoft, and also with ready made platforms like Elna. If you want to catch up on a year’s work, read this: I have been building AI agents for almost a year. Here are my experiences. and Personal AI x Agentic AI = Elna, a game-changer
About a year ago, I thought it was the “next big thing”, the future, the evolution of apps so advanced that they’d run your life for you. I was all in, nodding like I just found the key to the universe, until I took a deep dive into Christopher Noessel’s Designing Agentive Technology. Written in **2017.
Trust me, if your view of AI hasn’t been flipped upside down by this book, you’re doing it wrong.
Christopher Noessel is not just some random tech guy who posts “inspirational” AI takes on LinkedIn between influencer selfies. No, this guy is a heavy-hitter: He is a Design Principal at IBM with an AI focus, a UX design veteran, and the mastermind behind scifiinterfaces.com , and that is a place where sci-fi fans and design nerds unite to laugh about how the future actually turned out weirder than the movies. And this man knows AI like your grandmother knows her secret chicken soup recipe.
A quick recap on what Agentic AI is:
Agentic AI refers to artificial intelligence systems designed to act independently and proactively on behalf of users. Unlike traditional AI that waits around for commands, Agentic AI takes the initiative to assess situations, make decisions, and execute tasks based on its understanding of user needs, preferences, and context. Think of it as an AI that answers questions, and also can anticipate them, plans ahead, and carries out actions without you constantly having to look over its shoulder.
In practice, this means an Agentic AI could go beyond simply suggesting routes on a map. For instance, it could analyze your schedule, predict potential delays, and reschedule meetings if it senses that you’ll be late. When you want to buy stuff online, your Personal x Agentic AI does not show you a “buy” button, nah, it goes one step beyond, it purchases items that you’ve run out of or books services it knows that you will need. The technology wants to shift digital interactions from task-focused experiences to smooth and supportive partnerships, where the AI does much of the cognitive, creative, and logistical stuff.
And now a bit about the architecture of Agentic AI:
Agentic AI is not your run-of-the-mill, wait-for-a-command type of assistant…
It will become your friend that remembers everything and starts planning your life for you (whether you asked for it or not).
It runs on an impressive, and slightly intimidating architecture with memory, context-awareness, and autonomous workflow management to execute tasks so smooth that you’ll wonder if it’s reading your mind.
Let’s break down these components, but with a little more flair:
Memory: Regular reactive AI systems are like goldfish—stateless and forgetful. But Agentic AI? It’s more like an elephant crossed with your nosy neighbor: it never forgets. It has persistent memory structures that let it store contextual information over time, adapt based on past interactions, and build a disturbingly accurate profile of you. This isn’t just remembering your birthday; it recalls your past travel routes, favorite coffee stops, and which seat on the plane you prefer so it can suggest the perfect itinerary before you even say, “Siri, book a flight.”
Workflow Management: Think of Agentic AI’s workflow engines as a team of tireless minions juggling flaming swords. They coordinate tasks either in parallel or in a meticulous sequence, depending on what you need. Using decision trees, finite state machines, or probabilistic models that sound like they were named by a sleep-deprived mathematician, Agentic AI runs on a “sense-think-do” loop. Every phase of a task—whether it’s collecting data, making a decision, or actually doing the thing—is handled autonomously. It’s like a digital assistant that won’t bother you with “Should I send this email now?” but just handles it while you’re blissfully unaware.
Autonomy and Proactivity: The secret sauce of Agentic AI is its proactive nature. It’s like the coworker who anticipates your every need, minus the passive-aggressive post-its. Event-driven architectures and triggers allow it to react to changes in context, like your erratic search history or sudden weather shifts. Say an email with “meeting” in the subject hits your inbox—this AI won’t just highlight it; it’ll schedule the event, check the weather to make sure you don’t get drenched, and set a reminder that’ll actually get you there on time. It’s practically your digital mom, minus the guilt trips.
Task Management and Prioritization: Agentic AI is built with advanced algorithms that prioritize tasks like a Type-A overachiever. Using weighted scoring methods or reinforcement learning (basically, a self-improving to-do list on steroids), it decides which actions to execute first. And it’s got workflow orchestration tools to make sure no task is jumping the line or being done without the right info. Think of it as the project manager who actually knows what they’re doing—no sticky notes or coffee-fueled rants necessary.
Context-Aware Intelligence: For Agentic AI to blend in and feel human-like, it needs to process real-time contextual information faster than you can say, “Why is my smart speaker talking to me?” This is achieved using a cocktail of sensors, APIs, and data streams that all feed into the AI’s memory and decision-making algorithms. Imagine an AI managing your smart home like an ultra-efficient butler: it adjusts the thermostat, dims the lights, and locks the doors based on where you are, what time it is, and whether you’re likely to come home with takeout or just emotional baggage.
If you want to know more about how, say Google is planning on implementing Agentic AI through it’s product range - check out my next article on project Astra.
A vision for Agentic UX
In the last few weeks alone, it’s been all the AI leaders can talk about. OpenAI’s Sam Altman called agents AI’s?“killer function.” ?Google’s Demis Hassabis called agents?“the next big step change.” ?Anthropic’s Dario Amodei called agents the next?“significant unlock” ?and predicted that we are only three to 18 months away from this future.
And now for the vision part….
To me, it is the job of the Agentic designer to walk with confidence into the future of digital experience, and waving its metaphorical arms and shouting, “Behold, what is next, ye n@@bs”! But, let’s be real, we need to build a collective sixth sense for what this even looks like so we don’t end up with tech that’s as useful as a Casio with Wi-Fi.
The last time we shook up UX , shifting from the physical to the digital, selling it to business leaders was a walk in the park because “the medium was the message,” and everyone could nod and say, “Ah, yes, the web experience. I can see it: awkward fonts, broken links, and all.” When we threw around terms like “web experience” or “mobile experience,” it sparked vivid, if sometimes glitchy, images. The pitch practically made itself.
Fast forward to today, and saying “AI experience” or “AI-powered” lands with the clarity of a foghorn in a pillow factory. It’s too broad, too misunderstood, and honestly, makes most people’s eyes glaze over faster than my dog at a vegan BBQ. On the other hand, narrowing it down to “conversational interfaces” or whatever the latest trendy interaction pattern is doesn’t cut it either. It’s like describing a Ferrari as “a car with seats” ….. it’s true, but criminally underselling it.
I think Kwame Nyanning hit the jackpot when he coined ‘Agentic UX.’ It’s specific, it’s bold, and it finally makes us sound like we know what we’re talking about. More importantly, it aligns perfectly with the tech’s actual trajectory you know, the one where it’s evolving faster than your browser’s update cycle.
Here’s the thing about this new paradigm: it’s shockingly straightforward. In the old-school digital world, we built clunky scaffolding that held users up but made them do the heavy lifting — thinking, planning, navigating, and swearing when they clicked the wrong link. In Agentic UX, users aren’t on a solo adventure armed with nothing but hope and a scroll wheel. Instead, they have a built-in AI agent working overtime like the universe’s most loyal sidekick, turning every interaction from a lonely chore to a team effort with a partner that doesn’t bail when things get tough.
Imagine your average customer journey, now with an AI agent subtly supporting every step like an unsung hero. Picture it as a little layer underlining your digital trek, extending like a horizontal safety net ready to catch your confusion before you even trip over it.
In Agentic UX, this AI isn’t just there for moral support; it’s the muscle behind the magic. It flexes across three main types of weight:
From a UX perspective, it doesn’t really matter whether this AI agent struts in with a neon “Agent” badge or hides as a collection of subtle, unbranded features. The point is, as long as users feel supported and not abandoned in a digital wasteland, we’re on the right track.
Early examples of the agentic future of UX are already appearing today.?Adobe’s Gen Studio ,?Intercom’s Copilot , and?Dovetail’s Magic Experience ?are some of my favorite windows into the future of digital experiences.
Take Dovetail’s 2024 agentic UX suite, labeled “Magic” features. Dovetail, for the uninitiated, is a SaaS tool for researchers who spend their days uploading and analyzing interview data while mumbling, “Why did I choose this career?” Traditionally, the software would provide the structure for users to slog through their tasks. Now? It’s basically a research assistant with none of the awkward small talk, helping with transcript summaries, highlights, and clustering suggestions. It doesn’t even ask for coffee breaks.
In this early version of agentic UX, the AI is still more Watson than Jarvis — assisting but not taking full control. But as technology and training data get beefier, expect these digital teammates to bulk up their capabilities. Soon, you’ll go from thinking, “Wow, this AI is helpful,” to, “Why do I even bother showing up to work?”
Agentic vs Agentive vs Assistive
Agentive, agentic, assistive… I know. When you’re already lost, don’t worry.
These terms are the way that the tech industry wants to make you feel like that you’re failing a vocab test. But let me shed some light into this darkness for you:
Agentic is that behind-the-scenes tech wizard, which is quietly pulling strings. Agentive is that’s the suave lead actor taking control of your life on your behalf while you sit back with a drink in hand. Assistive.. yeah now that’s a very active one. It’s the faithful assistant that corrects your typos before you embarrass yourself in a group chat.
Agentive UX is about moving from “Look, Mom, I did a task”! to “Wow, I didn’t even have to do the task”! Think of an AI-powered camera that takes photos every 30 seconds while it is clipped to your shirt (good idea, GoPro? Just pay me a few thousand bucks through my Coffee page, see below). Now, at the end of the day, it sorts through the photos, picks out the best ones, and deletes the ones where your wife is mid-sneeze. Because the AI knows that women (and men) want to shine on a photo. It’s like Google Photos, but with the ruthless judgment of a reality TV show editor. Noessel’s big message was that you don’t design for tasks, but you design for the goals. Instead of, “Take photos”, think, “Create vacation memories without having to shout , ‘SMILE OR ELSE!’”
Agents are persistent loops
Forget input-process-output. Man, now that is some old-school stuff i used in the early 2000’s.
The Agents operate on a “sense-think-do” loop, which sounds way more fancer until you realize your dog (my dachshund) does the same thing when it hears the can opener. These agents sense inputs (facial recognition, biometrics, your desperate look at the coffee machine), think by processing that data (cue algorithms humming along like a caffeinated beehive), and do by taking action (screens, notifications, that overly perky robot voice).
And triggers?
Oh, they aren’t just 3 p.m. reminders. No, I am talking data-based cues that are so nuanced they’d make Mata Hari jealous. Say, an AI is noticing your digital trail of “10 ways to quit my job dramatically” and it gently suggests a list of therapists instead. Or how about an agent that waters your dog’s bowl when the temperature hits 20°C but caps it at twice a day because, well, Fido’s not entering a water-chugging contest.
More about Sense-Think-Do
So, how do agents think?
They rely on a “sense-think-do” loop that’s a step up from the standard human approach of “sense-guess-panic”. They sense with tech like biometrics, facial recognition, and keywords whispered under your breath. They think with algorithms that are as fast as a chef at a hibachi grill. And then, they do stuff like popping notifications, firing off messages, or controlling your smart fridge, which may or may not try to restock your ice cream without asking first.
Intermezzo: Why do we (tech writers et. al.) always use a smart fridge as an example?
Christopher Noessel tells us that understanding the user’s goals is critical. This means either watching users more closely than my nosy neighbor or making them set up rules so specific they feel like you are training a digital puppy. Example: “If the temperature hits 20°C, fill the dog’s water bowl, but only twice, and ignore the sad puppy eyes”. That kinda stuff. You get the idea.
Dr. H. Shyam Sundar at Penn State studied how perceived agency impacts user trust. Turns out that users just love predictive AI ….. until it messes up and suggests a smoothie when you’re searching for pizza. Dr. Anne McCarthy of the University of Glasgow explored how designing agentic personas can make tech feel more relatable and less like a passive-aggressive roommate. And MIT’s Dr. Pattie Maes explored context-aware AI, showing that we’re closer to having machines that actually understand us than we are to our smart speaker randomly blasting Bohemian Rhapsody at 3 a.m. because it misheard “bedtime”.
Agency is Proactive Intelligence
Here’s the litmus test for agentive UX:
Does your tech make users say, “Oh wow, that’s exactly what I wanted” or does it make them go, “Why did my smart vacuum just shouted slurs at me (read: Hackers took over robovacs to chase pets and yell slurs )”.
Agents should be proactive but not intrusive.
They need to predict, optimize, advise, and sometimes have the decency to know when not to act, like when your smart assistant understands that you are in a meeting and doesn’t break in with a not so subtle, “It’s been a while since you’ve urinated”!
Examples of proactive agency are everywhere: the Nest thermostat that learns your habits and preheats your house like it’s trying out for The Great AI Bake Off, or Waze, which of course map your route AND predicts that you will want to avoid that highway because there’s a crash involving a truck full of rubber ducks. Successful agents are like that one waiter who remembers your order but also knows not to recommend the “Chef’s Surprise” after last time you were in and was in the bathroom for a half hour.
Now there’s Jony Ive, who has crafted tech designs smoother than a Sinatra ballad, and she says that creating agentive experiences is like designing a butler: polite, insightful, and occasionally ready with an eyebrow raise. And of course, our resident commenter from the sidelines, Andrew Ng, the AI guru, believes that agentive tech is a step towards letting people focus on creativity while the tech handles all the soul-sucking stuff. And of course we want to hear what Tim Berners-Lee, the godfather of the web, has to say about it. But he issues a warning that we should design responsibly so that these agentive systems don’t turn into the next HAL 9000, and refuse to open the pod bay doors.
The spectrum of Auto vs Manual
The dream is to give users a choice between full auto mode (complete trust in your digital butler) and manual override (suspiciously side-eyeing your tech while muttering, “I’ve got this”). For this to work, mutual predictability is key. Users should know when and why their agents act, and agents should have enough context to stop themselves from adding “baby carrots” to your shopping list when you really meant “snack-sized candy bars.”
Think of a smart car with adaptive cruise control. One moment, you’re hands-free and (quietly) singing a happy tune, and the next, you’re wrestling back control because the car decided to brake for an imaginary squirrel on cocaine. It’s all about designing agents that handle your life without fuzz, and that allows you to switch from laid-back “cruise mode” to “get-outta-my-way” at a moment’s notice.
The future is not about designing apps that wait for input like wallflowers. It’s about creating proactive, trustworthy agents that say, “I’ve got this” but know when to step back and let you take the control. The line between creepy and convenient is up to us to design.
If your tech doesn’t have even a whiff of agentive UX yet, good luck, you’re already playing catch-up. Time to build smarter agents, because the future is watching you and wondering why you still use “password123”.
Signing off - Marco
Well, that's a wrap for today. Tomorrow, I'll have a fresh episode of TechTonic Shifts for you. If you enjoy my writing and want to support my work, feel free to buy me a coffee ??
Think a friend would enjoy this too? Share the newsletter and let them join the conversation. LinkedIn appreciates your likes by making my articles available to more readers.
Earlier articles on Agentic AI