The AI-native builder
A two-host podcast for designers becoming AI-native builders — part product manager, part UX designer, part engineer. One concept per episode, explained properly, with what it means when you sit down to build.
The autocomplete that ate the world
Next-token prediction, tokens, the context window, and why outputs are a distribution, not a result.
Listen or watch →
How LLMs actually behave
The intuition you need to design around a model: what it is doing, what it sees, and why it is never quite the same twice.
The autocomplete that ate the world
Next-token prediction, tokens, the context window, and why outputs are a distribution, not a result.
Brilliant and confidently wrong
What models are great at, how they fail — hallucination, sycophancy, lost-in-the-middle — and how to design for both.
Big, fast, or cheap
Model tiers, latency, and cost per task as product decisions, not engineering details.
Context engineering
The new information architecture: deciding what the model sees, when, and in what shape.
Prompts are interfaces
System prompts, instructions, and examples — designing for a reader who takes everything literally.
Getting JSON back
Structured output: turning prose into data your UI can render.
Retrieval: giving the model a library card
RAG, embeddings, and vector search — fetching the right knowledge at the right moment.
When the model presses buttons
Tool use and function calling: how AI goes from chatting to doing.
What the model remembers
Memory, state, and sessions — what persists, where, and who controls it.
Agents and orchestration
Loops, protocols, and teams of models — and the autonomy question at the heart of all of it.
Think, act, look, repeat
The agent loop, and why it changes what a 'feature' is.
USB-C for AI
MCP: the standard way to plug models into tools and data.
The autonomy dial
Multi-agent patterns, routing, verification — and when the AI should ask first.
Evals
The most neglected skill in AI product work: turning 'is it good?' into something you can test.
Your eval suite is your spec
Golden datasets, pass/fail criteria, and shipping on evidence instead of vibes.
Judges and humans
LLM-as-judge, human review loops, and measuring what matters.
AI-native UX patterns
Where your design background compounds: uncertainty, latency, trust, and control.
Designing for maybe
Confidence, citations, showing work, and calibrating trust.
Waiting, streaming, generating
Latency patterns, streaming, and interfaces the model assembles at runtime.
Undo for AI
Human-in-the-loop: previews, approvals, edits, and recovery.
Enough engineering to ship
The stack, the APIs, and the real multiplier: working well with coding agents.
The builder's stack
Web basics, React, a database, deployment, and calling an LLM API safely.
Pairing with a coding agent
Specs, small verifiable steps, and reading code well enough to review it.
AI product judgment
When not to use AI, what it costs, and how to keep users safe.
Should this even be AI?
Deterministic vs. generative, unit economics, and data flywheels.
Trust, safety, and the injected prompt
Prompt injection, privacy, and deciding what the model is allowed to do.