37 explainers
- What is a customer profile?
A customer profile is the data record of one customer that systems read. What goes in one, how it is built and kept current, and where profiles are stored.
- What is a feature store?
What a feature store is, its parts, the problems it solves, and how the main options compare, from Feast and Tecton to the cloud platforms.
- What is agent memory?
Agent memory is what an AI agent keeps and reads back across steps and sessions. The types, how recall works, and how Mem0, Zep, Letta and others compare.
- What is generative UI?
Generative UI is an interface assembled for each request from data and rules or a model. How it works, examples, and how it compares with A/B testing.
- What is propensity modeling?
Propensity modeling predicts how likely a customer is to take an action, like buying or churning, from labeled history and a trained model.
- Behavioral vs declared customer data
Declared data is what a customer tells you. Behavioral data is what they do. How they compare, map to zero-party data, and combine in one profile.
- Generative UI vs server-driven UI
Server-driven UI sends a screen description for the client to draw. Generative UI decides that screen per visit. How they differ and how they combine.
- Online vs offline features
Offline features hold history for training. Online features serve the latest value at inference. How the two stores differ and why they drift apart.
- Short-term vs long-term agent memory
Short-term memory in AI agents holds the current conversation; long-term memory lasts across sessions. How LangGraph, CoALA and Letta split them.
- What is a propensity score?
A propensity score is the output of a propensity model: a number between zero and one estimating how likely a customer is to take one action.
- What is agentic context?
Agentic context is the structured picture of the current situation an AI agent reads before it acts, including what the user is doing right now.
- What is an adaptive experience?
An adaptive experience, or adaptive UI, changes within a visit in response to what the person does. How it differs from personalization, with examples.
- What is identity resolution?
Identity resolution decides which cookies, device ids, logins and CRM records belong to the same customer. How it works, step by step, with an example.
- What is real-time intent?
Real-time intent is what a customer's current behavior shows they are trying to do, read from this session rather than predicted from past history.
- What is training-serving skew?
Training-serving skew is when the features a model reads in production differ from the ones it trained on. The model quietly gets worse, and nothing errors.
- Context engineering vs prompt engineering
Context engineering vs prompt engineering: how they differ in timing, ownership, failure modes and testing, with one customer question handled both ways.
- Feature store vs data warehouse
How a feature store differs from a data warehouse, whether you need one if you have a warehouse, and what Snowflake, Databricks and BigQuery add for serving.
- Intent vs propensity
Intent is what a customer is doing now; propensity is a predicted probability from past behavior. They answer different questions and need different data.
- What is an identity graph?
An identity graph is the store of links between the identifiers one customer uses, so a system can look up every id that belongs to the same person.
- Why generative UI needs real-time state
Why generative UI needs real-time state, what goes wrong when it reads stale data, and the ways to supply live state, compared with their tradeoffs.
- Deterministic vs probabilistic identity matching
Deterministic matching links identifiers on shared data like a login. Probabilistic matching infers a link from patterns. Precise versus wider reach.
- json-render explained
json-render is Vercel's open source library for generative UI. How the catalog, spec, streaming and renderer work, and how it compares with A2UI and MCP Apps.
- Propensity score vs intent bucket
A propensity score is a continuous probability; an intent bucket is a named state from a finite set. The bucket is easier to read, act on and change.
- What is an online feature store?
What an online feature store is, the databases behind Feast, Tecton, Databricks and SageMaker, and why keeping its values current is the hard part.
- Why AI agents need real-time context
Real-time context for AI agents: what goes wrong without it, where it comes from, how sources compare on freshness, and an agent tool call that reads it.
- Classification vs ranking
Classification sorts a visitor into a finite set of states; ranking orders a large catalog. Intent is classification; a catalog is ranking. Use both.
- Generative UI vs static A/B testing
How generative UI compares with A/B tests, multivariate tests and contextual bandits, and how you still measure a page that is composed per visit.
- What is a customer-facing AI agent?
A customer-facing AI agent talks to or acts for your customers in support, shopping or onboarding. The main kinds, why they fail and what they need.
- What is feature freshness?
Feature freshness is how old a feature value is when a model reads it. How it is measured, how batch and streaming compare, and what staleness costs.
- What is identity stitching?
Identity stitching joins a customer's activity across logins, devices and expired cookies into one profile, so what they did under another id is not lost.
- Building generative UI with an AI agent
How to build generative UI with an AI agent in six steps, from component catalog and tool schema to live context, validation, fallback and a holdout.
- Do you need a feature store?
Who needs a feature store and who does not, the signs on each side, the options in between, and how to weigh building one against buying one.
- How agents act on live customer behavior
How AI agents act on live customer behavior: pull context when the customer asks, or act first when behavior crosses a threshold. The loop and guardrails.
- Why a customer profile needs to be real-time
A real-time customer profile updates as the customer acts. What goes stale in a batch profile, how the two compare, and when batch is fine.
- Why propensity models stall in production
Propensity models stall between a notebook and production: labeled data, a feature pipeline, validation, serving and train-serve skew before any score ships.
- What is context engineering?
Context engineering is deciding what an AI agent has in its context window at each step. What goes in, the main techniques, and how it fails.
- What is LLM memory?
LLM memory is how an application makes a stateless model remember. The context window, chat history, saved memories, retrieval and fine-tuning, compared.
