From a good one-off dish to a recipe the whole team can cook
An interactive masterclass on leveraging the power of AI for UX research, and learning the context engineering techniques you need at each stage to level up your research process. Built for researchers at every level.
Hosted by Head UX AI Chef (lol) Kamala Alcantara · Senior UX Researcher and Voracious Vibe Coder.
APPETIZER
Where we are as a team
A recap of our AI sentiment workshop, served with a look at how AI fits into each phase of UX research.
What our team told us
Three boards from Ali's workshop. The same idea showed up on almost every sticky note.
Thinking partner. Bias-checker. Synthesis helper. First draft maker. Mostly in one-off chats that disappear tomorrow.
📌 How we use AI today
We want AI that knows OUR users, OUR data, and OUR past research. Not generic answers about anyone, anywhere.
📌 How we want to use AI
Empathy. The craft of deep research. Talking to real people. Making users feel heard. That is ours. Full stop.
📌 What stays human
The pattern: almost everything you want comes down to giving AI better ingredients.
ZOOMING OUT · THE WORLD'S TAKE
It's not just us. Here's how the world feels about AI.
Our team's mix of hope and caution mirrors what researchers are seeing globally. A few findings worth knowing, fresh off the press.
67%
of people worldwide feel net positive about AI
Anthropic, 81,000 interviews across 159 countries
27%
say unreliability (hallucinations) is their top worry
Anthropic, 81,000 interviews
80%
of researchers now use AI regularly in their work
User Interviews, State of Synthetic Users 2026
🌗 The big theme: "light and shade"
Anthropic's study of 81,000 people found that hope and worry about AI don't split people into camps. They live inside the same person. The very things that make AI useful are the things people fear. Someone who loves AI for emotional support is three times more likely to also fear becoming dependent on it. People are excited and uneasy at once, and that's normal.
😮💨 UX is feeling "AI fatigue"
NN/g calls 2026 the year of AI fatigue. Designers and researchers are tired of being told they'll be replaced, and tired of AI sparkle slapped on everything. Their verdict: the winners treat AI as a tool that recedes into the background, and the human craft (taste, judgment, research-informed understanding) is exactly what stays valuable.
NN/g, State of UX 2026
🤖 On synthetic users: useful, but verify
When asked about AI-generated "synthetic users," 47% of researchers are skeptical and want more evidence before trusting them. Top concern (88%): the quality of the insights. The takeaway echoes our kitchen rule. AI can give directional signals, but real humans and a real researcher's judgment stay in the loop.
User Interviews, State of Synthetic Users 2026
Why this matters for us
The whole world is landing on the same lesson we are: AI is a powerful prep cook, but people want it grounded, honest, and kept in check by human judgment. That is exactly what context engineering and skills give you. Sources: Anthropic 81k interviews, NN/g State of UX 2026, User Interviews Synthetic Users.
AI AT EACH RESEARCH PHASE
Where AI fits in your work
What you can hand off, and what you always keep.
AI CAN HELP WITH
YOU ALWAYS BRING
🗂️ Planning
Draft discussion guides, research plans, screeners, survey questions
Research goals, stakeholder context, what questions actually matter
Being present. Building rapport. Knowing when to go off-script.
🧩 Synthesis
Grouping tags, pattern-spotting across lots of data, first-draft themes
The non-obvious insight. The "so what." The story that matters.
📊 Sharebacks
Format findings, draft slide copy, generate charts, check for gaps
Stakeholder judgment, what will land, making people care
MAIN DISH
What Claude is and how to feed it well
How Claude works, what it is good and bad at, what humans must bring, and how to prep your ingredients before you ask a question.
What Claude actually is
🧠
How it works (plain English)
Claude predicts what a helpful, accurate response looks like based on a huge amount of text it was trained on. It is great at patterns it has seen a lot. It is less reliable on rare, niche, or very recent things.
🔄
It starts fresh every time
Every new chat, Claude has no memory of you, your users, or your team. It only knows what you share in that chat. This is why what you put in the prompt matters so much.
AI capabilities and limitations
Each property sits on a spectrum. The further right you push it, the more you need to verify and compensate with your own judgment.
CAPABILITYLIMITATION
✨
Next token prediction
Where do AI answers come from?
Well-worn paths: summarize, reformat, explain common thingsNovel territory, sparse patterns, "true vs. sounds true"
🌐
Knowledge
What does AI actually know?
Frequent, recent-in-training, mainstream topicsRare, post-cutoff, niche, local, or contested topics
💾
Working memory
What is it paying attention to right now?
Material fits the chat, you supply contextVery long docs, expecting cross-session memory
🎛️
Steerability
How much are you in control?
Short, concrete instructions ("respond as a table")Long reasoning chains, abstract asks
HUMAN COMPETENCIES VS AI PROPERTIES
What we each bring to the table
AI has properties. Humans have competencies. You cannot delegate the competencies.
Human competencies — what YOU must bring
Delegation — knowing what to hand off and what to keep
Description — explaining your context, users, and goals clearly
Discernment — judging whether the output is actually right
Diligence — reviewing drafts, catching errors, signing off
Working memory — holds and connects everything in the current chat
Next token prediction — finds the most likely helpful response
Knowledge — broad training across many topics and formats
CONTEXT ENGINEERING
Chop before you cook
Context engineering is a fancy term for a simple habit. Give Claude the right information before you ask your question. Lay out your ingredients on the counter first.
THE PROMPT
"Why do parents cancel their Coterie diaper subscription?"
WHAT YOU GET
Generic bullet points that could describe any subscription product anywhere. Sounds plausible but is not grounded in your real subscribers. Claude fills in the gaps with guesses. Risk of hallucination.
THE PROMPT
"You are helping me analyze why parents are canceling their Coterie diaper subscription. Here is our research plan with the top study goals [paste]. Here are 8 transcripts from our recent 60-minute conversations with churned parents [paste]. Using ONLY these, what are the top 5 reasons for canceling, sorted by most frequent and most convergence across participants? Quote the source transcript for every point."
WHAT YOU GET
Patterns tied directly to your real subscriber behavior. Every point backed by a quote you can verify. Sorted by frequency AND convergence so the most important signals rise to the top. A real head start on your synthesis. Nothing invented.
The four-step prep checklist
1
Set the role
Tell Claude what it is helping you with. Example: "You are helping me analyze interviews about why parents cancel their Coterie subscription."
2
Share your stuff
Paste in real things. A persona. Some quotes. A past finding. Your actual evidence.
3
Add the guardrail
Say: "Use only what I gave you. Back up every point with a quote." This stops Claude from guessing.
4
Say what you want
Tell it the format. A table? A list of themes? A draft readout? Be specific about what the output should look like.
DESSERT
Skills and artifacts
What a skill is, how Claude uses it, and a live look at two real artifacts I built to supercharge two common UX processes.
What is a skill?
A skill is a saved set of instructions for Claude. Think of it like a recipe card. You write it once. Then you and your whole team can use it over and over without starting from scratch.
WITHOUT A SKILL
You retype the same prompt every time
Quality changes depending on who asks
If you leave, the method leaves too
New teammates have to guess how you did it
WITH A SKILL
One click. Scalable output.
Close to same quality no matter who runs it
The method lives with the team
New teammates can use it on day one
How Claude uses a skill
1
You write the skill
A document that tells Claude its role, what inputs to expect, what steps to follow, and what the output should look like.
2
You share your stuff
Open Claude, share the skill document, and paste in your actual work. Transcripts, a persona, rough notes.
3
Claude follows the recipe
Claude reads the skill and your materials, follows your steps, and produces a draft in the format you asked for.
4
You review and finish it
You are the chef. You check the output. You add your judgment. You decide what ships. Claude never sends anything on its own.
LIVE DEMOS · PLAY WITH THEM
Two artifacts from my kitchen to yours
Artifacts I built with Claude and use! (The HTML output is a remix of an existing artifact my team and I tweaked.) Go ahead and play with them right here, or open them full screen in Claude.
ARTIFACT 1 · LIVE
UX Research Plan Generator
Fill in your research details and this tool builds a complete, structured research plan as you go — updating live as you type. It walks you through everything: background and problem framing, research goals and questions, methodology, participant recruitment, assumptions, scope, timeline, and open questions. Add, remove, or reorder sections to fit your study. When you're done, export it as a PDF to share with your team.
An interactive experience that guides you through a full heuristic evaluation — whether it's your first time or your hundredth. The artifact walks you through each of Nielsen's heuristics, what to look for, and how to score issues with a built-in scoring guide to keep everyone calibrated. When you're done, download a beautiful HTML report you can share directly with your team. Includes an optional AI review layer for evaluating AI-powered or AI-assisted designs.
Practice context engineering and prompting with a real sous-chef
Pick a research phase, write a prompt, and hit "🥄 Taste test my prompt" for live feedback from Claude. It scores your four ingredients, suggests a rewrite, shows you a chef's version, and flags any UX artifact you should paste in to make it stronger. Five scenarios, one per phase of research.
COOK AT HOME
One thing to try this week
Pick your level. Every one of these moves you from one-off chats to repeatable, evidence-grounded work.
Pick a question you would ask AI anyway. Before you hit send, paste in a persona or a real quote. See what changes. That is it. One prompt. Notice the difference.
🍃 LINE COOK
Save a prompt you reuse
Find something you keep retyping. Add step 3, the guardrail: "Use only what I gave you. Back up every point with a quote." Save it somewhere you can grab it next week. A doc, a note, anywhere.
🌶️ HEAD CHEF
Write your first skill
Pick one thing you do often. Write the five parts of the recipe card: when to use it, what to bring, the steps, the guardrails, and what it should make. Ping Kamala and we will build it together.
🥡 TO-GO DISH
Remix an artifact to make it your own
Open one of the artifacts above in Claude, then ask Claude to change it. Make the Research Plan Generator match your team's template. Tweak the Heuristic Eval severity labels. You do not need to code. Just tell Claude what you want different. That is how you learn what these can do.
The recipe card — five things every skill needs
Screenshot this or download it to use as your guide.
1
When to use it
The trigger. Example: "Use this when tagging interview transcripts."
2
What to bring
The inputs it needs. Transcripts, a persona, a research goal, etc.
3
The steps
How YOU do this kind of analysis. Write that down. That IS the skill.
4
The guardrails
What should Claude never do? No making things up. No sharing without a human check.
5
What it should make
The finished output. A table, a doc, a set of themes with quotes?
AI preps the kitchen. We stay the chefs.
The empathy, the craft, the deep work? Still yours. That was never on the menu for AI.