AI Agent Experience (AX)
From Task Completion To Designing For Task Supervision.
Something fundamental just broke in product design.
You spent years learning that users want control. Click here, tap there, scroll down, fill out this form.
The entire discipline of UX rests on the assumption that people want to directly manipulate interfaces to get things done.
That assumption is not correct now.
AI agents don't need your carefully crafted user flows.
They don't care about your conversion funnels.
They're not going to click through your onboarding sequence.
They're going to accomplish user goals through paths you never designed, using tools you never intended them to use.
If you're still designing interfaces like it's 2019, you're solving the wrong problem.
What Are AI Agents and Why They Change Everything
Siri is not an agent. Alexa is not an agent. These are voice-activated search engines that respond to specific commands within narrow contexts. Real AI agents operate differently. They take complex, multi-step goals and figure out how to accomplish them.
You don't tell an agent to "book a flight." You tell it: "Find and book a round-trip flight to London for next month, balance cost and flight duration, add it to my calendar."
What happens next breaks every assumption about user interfaces.
Behind that single instruction, something unprecedented happens.
A primary agent spawns specialized sub-agents:
A Research Agent scrapes flight data from multiple airlines and aggregators.
A Planning Agent analyzes options against your preferences (avoiding red-eyes, preferring certain layovers).
A Booking Agent interfaces with airline APIs to secure the ticket.
A Communication Agent drafts booking summaries and handles notifications.
The user never clicks through a flight booking funnel. They never see the intermediate steps. They supervise a system of systems where outcomes aren't always predictable.
The agent might use tools you didn't design for it, or combine data in ways you never anticipated.
This is not an incremental improvement to existing interfaces. This is a completely different relationship between humans and software.
How to Design for Agents
If users aren't controlling every step anymore, what do we build? We build the cockpit. The navigation system. The dashboard that lets users supervise the journey. This is Agent Experience.
From Execution to Supervision
Users become managers instead of operators. They set goals, monitor progress, and approve, reject, or modify what the agent proposes.
Our interfaces must evolve from forms and buttons to dashboards and approval queues.
The most critical user flows are no longer about task execution.
They're about intervention and verification.
Language as Interface
In agentic systems, conversation replaces clicking. The prompt isn't just a query. It's the start of an ongoing dialogue.
How does an agent report progress? How does it ask for clarification? How does it present findings? These become core design problems.
An agent that communicates poorly will be trusted less, regardless of how well it performs technically.
Designing for Co-creation
The traditional handoff between design and engineering doesn't work anymore. Designers can't create mockups without understanding the AI's architecture, capabilities, and limitations.
What tools can the agent access? How reliable are they? What are the common failure modes?
This requires collaboration that goes deeper than typical agile workflows.
Product and engineering teams must define agent goals and user touchpoints together, as a unified system.
The Technical Reality
Building for AX means designing two interfaces simultaneously:
Human-to-Agent Interface This is where users oversee, approve, edit, and guide agent behavior. Think mission control, not task execution.
Agent-to-System Interface This is where agents interact with APIs, databases, and tools. This interface needs to be robust, well-documented, and optimized for machine efficiency.
Anticipating user needs changes completely. We must predict the most likely follow-up actions to agent outputs. If an agent drafts an email, the next steps are probably "send," "edit," or "save as draft." The UI must surface these actions contextually and efficiently.
Building successful agentic systems comes down to building trust. As we give more control to autonomous systems, our job is ensuring human users remain the ultimate authority. They need tools to confidently supervise, collaborate with, and trust their AI counterparts.
The era of Agent Experience is here. The question is whether you'll help define it or watch it happen to you.
Agent Experience in the Real World
Agent Experience isn’t a theory-it’s already shaping how modern tools behave, and how people interact with them on a daily basis.
In Notion, users aren’t typing from scratch anymore. They’re prompting the AI to summarize meeting notes, rewrite paragraphs, or brainstorm bullet points. The work is no longer about typing-it’s about reviewing, adjusting, and steering the outcome. The writing interface now lives inside a dialogue box.
Replit’s Ghostwriter flips the role of a developer. Instead of building line by line, devs are guiding a coding assistant through intent. When the AI suggests code structure or highlights bugs, the human is supervising-not executing every step.
This isn’t automation of tasks-it’s augmentation of thinking.
On Shopify, merchants use Sidekick to manage stores more intuitively. They don’t fill out forms to run promotions-they describe their goals in natural language. Sidekick figures out the rest. What matters now isn’t how well the merchant understands the UI, but how clearly the agent understands the merchant.
Salesforce’s CXO Kat Holmes explains that true AX means designing systems both for AI agents and human supervisors. For instance, an agent might receive a customer request—say, changing a delivery time for a large appliance. To complete the task, the agent must coordinate across user profiles, shipping systems, inventory, and delivery partners.
The real UX challenge isn’t just building an agent that executes; it’s building the orchestration layer that coordinates subtasks, surfaces uncertainties, and lets humans intervene where needed .
And these are just the early signals.
Takeaways
In each of these cases, product teams had to reframe the core experience. The end user isn’t driving every action anymore. They’re partnering with the system-watching, course-correcting, and intervening when needed.
We’re moving from designing for task completion to designing for task supervision.
In this new world, AX isn’t a feature. It’s a philosophy.
It forces us to rethink how products are built, how trust is earned, and how users collaborate with systems that think, act, and decide.
The interface isn’t the journey anymore-it’s the checkpoint. And the winner won't be the company that adds agents the fastest. It’ll be the one that builds the clearest path to human authority.
Because when everything becomes automated, designing for when-and how-humans step in becomes the most important design challenge of all.
This is an open edition of The Product Voyagers. To unlock full playbooks, tools, templates and expert takes — go Prime and get everything in your inbox.
📚Related Reads
Cobus Greyling stresses planning for human intervention as a user experience necessity, not just a safety net.
Wyndo captures the core of Agent Experience by watching an app come to life from a simple prompt, with no traditional coding—just direction, iteration, and feedback. The story highlights a broader shift: product creation is moving from hands-on construction to guiding intelligent systems that interpret and act on human intent.
If you're curious about AI agents and what really drives automation, grab a coffee and keep reading
 | A guest post by
|
Enjoyed this write up, thanks for putting it together!