AI
Shipping LLM features that survive contact with real users: evals, agents, retrieval and restraint.
Multimodal interfaces: beyond the chat box
When chat is the wrong interface: designing AI products with the right modality — voice, image, structured output, inline suggestions — and how to choose.
Teaching what the AI can do: feature discovery as curriculum
AI features fail when users don't know what to ask. Designing discovery surfaces — prompt starters, worked examples, capability demos — that teach, not decorate.
When to build a voice interface (and when a form wins)
A decision framework for voice UI: where speech beats typing, turn-taking and latency design, barge-in patterns, transcription honesty, and voice privacy.
Analytics for AI features: what to measure
Chat volume tells you nothing. The instrumentation stack we use for AI features: acceptance, correction signals, task completion, cost-quality overlays, weekly review.
Chunking is a design decision: RAG retrieval craft
Chunk sizing, structure-aware splitting, metadata, hybrid retrieval and reranking — the retrieval engineering that decides whether your RAG feature answers or shrugs.
The prompt box is your whole interface — design it like one
An empty prompt box teaches nothing. Placeholder pedagogy, suggestion chips, command discovery and attachment UX that make AI assistants learnable.
The model deprecation playbook nobody writes until it's 3am
Your LLM provider just announced a deprecation window. How to migrate without panic: abstraction layers, eval-gated cutovers, shadow traffic and honest customer comms.
Designing AI features users can trust
Trust is an interface property, not a policy page. Citations, confidence, correction affordances and honest capability copy — how we design AI people rely on.
Evals-first development: the build order that actually works
Write the golden set before the prompt, version prompts like code, and gate merges on eval scores. The workflow that makes LLM features shippable.
When the AI says no: moderation UX
Every generative AI feature eventually refuses a user, and most refusals are dead ends. Designing blocked, revised and appeal states that preserve dignity.
Onboarding users to an AI assistant
First-run design for copilots: capability discovery without homework, starter prompts that teach, mental-model calibration, and measuring adoption honestly.
When the model is wrong: designing failure states for AI features
AI failure is a design state, not an exception. Wrong-answer UX, graceful degradation tiers, feedback that improves the system, and copy that keeps trust intact.