AX Design: The shift from Doer to Orchestrator

‍

Back in the ancient days before the internet, design was flat, static, defined and confined by the physical. A designer decided (or was told) what was going to be on a page, put it on the page, then printed the page. It was a dictatorship - the audience saw what the designer wanted them to see. The advent of digital experiences, and User Experience (UX) design as a discipline drove a shift towards a more democratic relationship - designers needed to understand humans and their intent to ensure the design facilitated the desired outcome. A sort of choose-your-own-adventure where the user has some input and control over their journey. As the sands of time move on we find ourselves on the precipice of another shift, and the defining moment of a new direction in design. 

I’m referring to AX Design, or Agent Experience Design. We are so much at the cutting edge of this discussion that the design powers-that-be have not even settled on the precise meaning of the words. Currently, AX Design could be referring to the technical approach of making a product legible to a piece of software or agent such as Claude of ChatGPT: clean APIs, structured data, the plumbing that lets an AI agent read and operate a website or platform on its own. It can also be the role of designing the experience that determines what it's like for a person when an agent is doing some or all of the driving on their behalf, while mitigating the risks and enhancing the benefits of an AI experience. At Octave we work within both of these definitions of AX Design, but in this blog post I’m talking specifically about the human part, and how the designers here at Octave are transferring our existing knowledge, experience and expertise in Human Centred Design and UX Design into designing for humans interacting with agents.

To successfully adapt HCD and UXD into yet another acronym - AXD, we need to understand the differences between them, the changes in what we are designing and how we are designing it. Old UX was deterministic: the user acts, the system responds in a generally predictable way, and a designer can map most paths in advance. The user can choose-their-own-adventure, as long as it's one of the adventures the designer mapped out. An agent doesn't work like that. It can act on its own, and you can't always predict exactly what it'll do next. The experience shifts from choose-your-own-adventure to enabling the user to construct their own world, their own universe, an infinite number of adventures. Design shifts from making tangible artifacts that describe a multi-faceted linear journey to setting the foundations and guardrails so that the ways for a user to reach a prescribed outcome are limitless.

An example: Say you want to build an agentic experience for an educational scenario. This isn't hypothetical; it's something we're currently working through with one of our clients. The desired outcome is defined: ensure participants learn enough to do the required task or job, within an established context and location, safely, responsibly, and to a set standard.

What's different is the journey. What used to be watch this video, attend this class, read this book, sit this assessment, receive this accreditation can now be different for every individual learner. Theoretically, this improves the quality and efficiency of learning by catering to different learning types, needs and backgrounds, and reduces the overheads of creating, maintaining and teaching the learning materials.

There's so much to unpack in that scenario, and there will be in every agentic experience like it. How do we assess this kind of learning? Will an AI-led learning experience be thorough enough? Will learners actually want to engage this way? Will we still have a shared understanding of what's been learned, if everyone's learned it differently?

Part of the AX designer's job is exactly this: working through those questions and risks, setting the parameters of the world the user can create, and defining the guardrails so that any interaction and journey with the agent still brings every learner to the same outcome, able to consistently apply what they've learned, to a set standard, within an established societal and workplace framework.

So far, the guardrails we’ve worked with are a mixture of fixed and flexible. Flexible guardrails bend to the individual learner, like language, pace, or format. Fixed defines things like the safety standard, the accreditation criteria, the thing that can never move no matter who's learning or how.

For example, a fixed guardrail with a flexible allowance might be something like:

“The outcome for the learner is to be able to perform the required task in a safe, responsible manner within an English-language, New Zealand-based workplace. A learner may, however, use translation and foreign-language tools and learning aids to move them toward the desired outcome.”

This is a guardrail that lets a learner whose first language isn't English engage in a way that suits their own learning needs, while still landing on the same outcome as everyone else. It’s guardrails like this that will be a designer's superpower. Traditionally, the advocate of the user, the defender of disparate mental models, the conduit between intent and code, a designer can now construct a system that moves a user to the desired outcome without compromising an individualised experience that is tailored to the user in every way.

Going beyond a non-linear user experience, agents are expected to learn, remember, and improve. Designers will not only be designing a system, but they’ll also be setting the parameters for a kind of agent/user relationship. To use our educational example again, learning doesn't happen in one sitting. A learning agent will remember what a learner has already tried, what's worked and what hasn't, and adapt the next step accordingly, rather than starting from zero every time someone logs back in. For those sceptics among us - designing the rules of engagement and limits of the relationship will be absolutely key in ensuring agents do not overstep and quietly embrace the position of robot overlords.

Everything is different, but some things stay the same

For all that's new, an AX designer is still dealing in the human side of things. We still have to consider how the human interacts with the agent. Interface design doesn't disappear so much as stop being pre-built: instead of a designer laying out fixed screens in advance, the agent increasingly generates or requests the interface it needs in the moment. Where interface still exists, it needs to be far more fluid and flexible than anything we've designed before.

At Octave, there are a few things that stay the same:

• We always start with the why. What problem are we actually trying to solve? This matters even more now, given the vastness of options AI opens up to us. Without first asking what specific problem we're solving for the user, we're just adding to a graveyard of lonely, unused, un-useful and possibly disgruntled AI agents.
• We deal in user intent and user experience, not colours and buttons. Understanding how humans think, operate and interact still trumps flashy visual design, though it doesn't hurt for something to be both useful and beautiful. 
• We understand and clarify the goals, opportunities and limits of the organisations we work with. A good AI idea doesn't solve anything if the internal understanding, governance and technical foundations aren't in place. In the learning example above, the guardrail only works because someone has already done the hard job of deciding what the organisation will and won't accept as "learned." 

‍Where this leaves us...

The educational scenario we are currently working on here at Octave is a small window into a big shift. Every agentic experience we design from here starts with the same questions we always ask: why are we building this, what does the person actually need, and is the organisation actually ready? The technology and tools might be different, but underneath it all the job hasn't changed: understand the human, set them up to succeed, and get out of the way.