AI Native
Build products where the model is the interface, and the output has a shape.
Sessions
Go in order. Each session is one idea you can drag, step and break.
01AI features in products
2 live · 5 comingThe building blocks of an AI feature.
Already live in other tracks: Embeddings and similarity search
Interactive
LiveForce valid JSON
Turn a schema on and watch illegal tokens get refused.
Open session →
Interactive
LiveCall a tool, then answer
Give the model a weather tool, then take it away and watch it guess.
Open session →
Interactive
Coming soonClassification and routing
Label a ticket, an email or a lead with a model instead of rules.
Interactive
Coming soonExtracting data from documents
Turn invoices, resumes and contracts into clean fields.
Interactive
Coming soonSummarization that keeps the facts
Shorter, without dropping or inventing details.
Interactive
Coming soonWriting and rewriting features
Drafts, tone changes and translation inside your product.
Interactive
Coming soonSearch that understands meaning
Users type what they mean, not the exact words.
02AI user experience
2 live · 5 comingDesigning for a slow, sometimes wrong collaborator.
Interactive
LiveStreaming UI and partial results
Show the answer as it arrives, and handle half-finished output.
Open session →
Interactive
LiveDesigning for slow responses
Progress, skeletons and doing work before the user asks.
Open session →
Interactive
Coming soonShowing sources and confidence
Citations and uncertainty that help users decide what to trust.
Interactive
Coming soonUndo, edit and regenerate
Keep the user in charge of what the AI changes.
Interactive
Coming soonChat vs embedded AI
A chat box, or AI built into buttons and forms people already use.
Interactive
Coming soonCopilots and inline suggestions
Suggest in place and let the user accept, edit or ignore.
Interactive
Coming soonWhen the model is wrong
Fail in ways users notice and can recover from.
03Shipping AI features
2 live · 4 comingFrom a demo to something people rely on.
Already live in other tracks: Evals: is the new prompt better?
Interactive
Coming soonFrom prototype to production
What breaks between a great demo and real traffic.
Interactive
Coming soonEvals for one feature
A small test set per feature, run on every prompt change.
Interactive
LiveCost per feature and per user
Price an AI feature before it surprises the bill.
Open session →
Interactive
LiveA/B testing AI features
Compare prompts or models on real users safely.
Open session →
Interactive
Coming soonCollecting user feedback
Thumbs, edits and abandons as signals for the next version.
Interactive
Coming soonUpgrading the model under a live feature
Switch models without changing what users get.
04AI in your development workflow
0 live · 4 comingBuilding software with AI as a teammate.
Already live in other tracks: A coding agent's loop
Interactive
Coming soonPair programming with AI
Where AI speeds you up, and where it quietly slows you down.
Interactive
Coming soonWriting specs for agents
Describe the task so an agent can finish it without guessing.
Interactive
Coming soonAI code review
Let a model review pull requests, and check its findings.
Interactive
Coming soonGenerating tests with AI
Useful tests, not ones that only confirm the code as written.
05Case studies: build an AI feature
0 live · 4 comingOne feature, end to end.
Interactive
Coming soonA support reply assistant
Draft replies from your help docs, with sources and an approve button.
Interactive
Coming soonAn invoice extractor
PDF in, validated JSON out, with low-confidence fields flagged.
Interactive
Coming soonMeeting notes and action items
Transcribe, summarize and assign follow-ups.
Interactive
Coming soonSmart search for your product
Meaning-based search with filters and fallbacks.
In production
Tools you can use
20 technologies across 6 live sessions. Same names in any language — pick a session to see how the idea maps to the real world.
Services, platforms, protocols and standards — not one stack.
Schemas constrain generation so parsers never see invalid shapes — across vendors and runtimes.
- JSON SchemaStandard
Describe object shape; many validators and model APIs accept it.
Official site ↗ - OpenAI Structured OutputsService
Response format enforced to a JSON schema.
Documentation ↗ - Google Gemini response schemaService
Constrained JSON from Gemini models.
Documentation ↗ - Outlines / GuidanceSoftware
Open-source constrained decoding for local models.
GitHub ↗
Models emit structured call requests; your app runs the tool and returns results — same pattern everywhere.
- OpenAI function callingService
Tools defined in the API; model returns JSON arguments.
Documentation ↗ - Anthropic tool useService
Claude requests tools with typed inputs.
Documentation ↗ - Model Context Protocol (MCP)Protocol
Standard way to expose tools to agents and IDEs.
Official site ↗ - OpenAPI toolsStandard
Describe HTTP tools once; many agents can call them.
Official site ↗
AI SDKs ship partial-object streaming helpers.
Design systems for AI products ship these patterns.
Every provider bills by tokens; observability tools track it.
Roadmap
Coming soonA step-by-step path through the track, with sessions and practice at each stage.
- Stage 1
AI features in products
- Stage 2
AI user experience
- Stage 3
Shipping and measuring
- Stage 4
AI in your own workflow
Projects
Coming soonBuild something real to lock in the skill.
- A support reply that only returns JSON
- An agent with two tools
- An eval set of 20 questions