Design Patterns for AI Interfaces (10h video + live UX training)

Get the course! 10h video + UX training
10h video + live UX training
30 video lessons + live UX training.For senior designers, UX researchers and design leads.
AI doesn’t have to be slow, verbose and unreliable. We can shape AI experiences way beyond a chat — and guide users to better prompts and results without typing. This course helps with just that.
You’ll learn actionable design patterns for AI, real-life examples and AI research, how to lead conversations with engineers and PMs — and how to design realistic and feasible AI experiences that users trust, love and use.
Jump to the table of contents ↓
What can we design with AI today? We’ll study the different flavors of AI UX, the emergence of AI fatigue and how people actually use AI. We’ll also touch upon product-market-fit gap, quiet vs. visible AI, AI search optimization — and why "AI-second" products might work better than "AI-first" products.
How magical is AI really? Well, let’s take a closer look at what actually happens inside of that magical box. We’ll break down confusing AI vocabulary to have meaningful conversations with AI engineers and product managers.
That’s where our UX journey begins! As designers, we can shape and influence how AI responds to user’s queries. It all starts with system and user prompts — and custom instructions written by users.
For an intelligence of any kind, AI is remarkably forgetful — especially during longer conversations with users. It’s not surprising because it actually has a limited “memory”, or “context window”.
Well, it’s uncharted waters now! Let’s explore temperature, top_p, top_k and max_tokens — slightly mysterious but very useful AI parameters that control how “creative”, predictable and verbose AI's responses are.
We often hear about AI having “memory”, but how does it actually work? We’ll explore how AI mimics memory, the critical UX challenges that memory management creates and why we need to set up different personas and give users control over what the AI remembers about them.
As humans, we aren’t very good at articulating our intent well. So why do most AI products place the burden of crafting the perfect prompt on users' shoulders?
Let’s see how we can change that, and why why a good old-fashioned UI with nudges and controls is so much needed to complement an empty text box.
Over the years, AI experiences have changed. Compared to the early days, AI today spends way more time planning, researching, “reasoning”, asking for more context and stitching data together. For some tasks users need slow AI, and for others AI must deliver results faster, despite its drawbacks.
The most significant attribute of AI agents is autonomy. But how much control are we willing to give AI agents to take over and do our work for us? We’ll look at how AI agents work, where they struggle, how they are organized and why it’s necessary to keep AI agents on a leash.
AI has inherent flaws and limitations. As designers, we need to know these constraints to design viable and feasible AI experiences. We’ll explore AI naivety and hallucinations, slow speed and low accuracy, forgetfulness and AI model collapse.
No technology is free — it always has some costs attached to it. In fact, AI has an enormous environmental cost, from massive energy consumption and toxic mining to carbon emissions and water required for cooling.
How do people actually use AI? How do AI experiences fit into their daily workflows? And what do we know about common user behavior patterns and shortcomings of AI experiences today?
For any AI feature to deliver value, it must first be discovered — and that’s where the troubles begin. Prompt suggestions often go unnoticed, “sparkles” are often misunderstood, and even prominent AI assistants get ignored.
To many, an empty text box is remarkably scary. It suggests to “Ask me anything”, but then users don’t really know what to ask, and what format, and at which level of detail.
Let’s see how we can help users understand what AI can do for them — e.g. with task-oriented suggestions, good old-fashioned filters and UI controls.
Context awareness is knowing what it remembers and why it generated a specific response. Making AI’s context transparent creates trust, but it can go way beyond showing sources.
Let’s see how we can show scope with context chips, how to map statements to specific text fragments and give users better overview of what AI knows about them.
Key AI features are often left between the lines, inviting assumptions and guesses. And typically that’s the job of onboarding.
Let’s see how we can design better onboarding for AI features, from pre-prompts and interaction modes to custom user personas and recipe guides.
The time has come! Let’s dive into the myriad of practical design patterns for AI interfaces. Let’s explore how we can address them — from scaffolding and interaction to input UX, output UX, refinement and orchestration.
So where should AI actually live in your product? Let’s dive into “scaffolding” — the layout, composition and integration of our AI features.
We’ll explore center-stage experiences, collapsible widgets, side panels and complex flow builders. We’ll see how AI can go way beyond a text box.
We’ll drill down into very impactful but underutilized patterns for scaffolding. We’ll explore inline overlays, infinite canvases, AI-powered data grids, third-party systems and the rise of the AI-enhanced search.
We often think about chat input, but there are different interaction modes — from the familiar text and voice inputs to visual, haptic and ambient interactions.
And then, we also have multi-modal experiences, and we can even turn a simple doodle into a working prototype.
Designing an AI-human chat is an entirely different beast than designing a social messaging experience like WhatsApp or Telegram.
People are very good at spotting an AI disguised as a “human assistant”. We’ll explore design guidelines for transparent, useful and respectful human-AI interactions.
F