Key answer: AX (AI Experience) is the design of experiences where users collaborate with artificial intelligence – in-app assistants, conversational interfaces, intelligent automations. The discipline's core is designing trust – clear boundaries of what the AI can do, honest communication of uncertainty, and easy human takeover of control.
Table of contents
The short answer
AX (AI Experience) is the design of experiences where artificial intelligence is part of the product: in-app assistants, conversational search, automatic summaries, smart suggestions. What makes AX a discipline of its own fits in one word: uncertainty. A classic interface is deterministic – the same click gives the same result. AI can be brilliant and can be wrong. Designing that duality is AX.
Why the old patterns fall short
For decades we designed machines that execute commands: the "save" button saves, always the same way. Language models brought something new into products – a system that interprets, proposes and sometimes confabulates. The old patterns have no answers for it: what do you show when the system "isn't sure"? How do you design a field that accepts anything? What happens when the user receives a wrong answer delivered in a confident tone?
Principles of good AX
1. Design boundaries, not magic. Users must understand what the AI can and cannot do. Good products state the scope plainly (example prompts, capability descriptions) instead of feigning omniscience and disappointing.
2. Communicate uncertainty honestly. A "90% confident" answer and a guess should look different: sources next to facts, "AI-generated" labels, easy verification. Trust is built with candour, not a confident tone.
3. The human can always take the wheel. Every AI action should be previewable before execution, reversible after, and editable rather than take-it-or-leave-it. Automation without a brake is a liability.
4. Input is design too. An empty chat box paralyses. Good AX designs the input: contextual suggestions, prompt templates, clarifying follow-up questions, the ability to refine a result instead of starting over.
5. AI in the background often beats chat. The best AI implementations frequently have no conversation UI at all: automatic categorisation, in-context suggestions, a summary at the top of a document. Chat is one pattern, not the goal.
Where to start in your product
Not with the technology, but with an inventory of the user's work: where do they spend time on tedious, repetitive, language-shaped tasks? Writing replies, summarising, searching by meaning, categorising – natural candidates. Then a prototype on a small slice, user testing (yes, AX gets tested too), measurement – and only then scaling.
The common ordering mistake: "we bought model access – what shall we add?". Reverse it: "what frustrates our users, and is AI actually good at it?".
What this means for companies – now
Conversational and agentic interfaces are no longer a curiosity: users are learning that software can be talked to, and they expect intelligent help from products. Companies that think AX through today will build an advantage that is hard to copy, because good AX is dozens of design decisions across the whole product.
At Sweet Lava we design AX as part of complete product design – from strategy through prototypes to implementation. See how we work with AI in products.