Featured essay · Oct 26, 2026
Put conversation practice inside your product
Our first public API brings authored rooms, a cast of up to five from 52 AI characters, a Coach in every turn and saved practice into your application.
Read the essayArticles
Ideas tested against the work.
Essays on participation, continuity, product character, responsible technology, and the choices beneath a family of distinct experiences.
The library
Design notes with enough room to be useful.
product design · Oct 5, 2026
AI is a great storyteller. It is not a good story writer.
We spent most of a year building a game that depends on AI telling a story every turn. We came out believing the opposite of what everyone fears about AI and writers.
Read articleproduct design · Oct 5, 2026
Games that make you think
Some fun asks nothing of you. Some asks you to read closely, form a theory, test it, and earn the next door. Adventures is built for the second kind.
Read articleproduct design · Oct 5, 2026
Making an adventure: from a story idea to a world you can play
How an Adventure is written, built and tested: what people decide, what the AI storyteller is given, and what we are learning writing the second one.
Read articleproduct design · Oct 5, 2026
The first date goes quiet. Give the conversation something to work with.
Step into a first-date practice with Conversations: a specific answer, a shared tangent, and coaching about what to try next.
Read articleproduct design · Oct 5, 2026
The storyteller holds no story
Why an AI storyteller's persona contains none of the plot it tells, the mistake that taught us, and the three rules that replaced it.
Read articleproduct design · Oct 5, 2026
When a friend is grieving, a pause can be part of the conversation.
See a Conversations demonstration in which the learner chooses silence, the fictional friend responds, and the debrief names a useful growth edge.
Read articleproduct design · Oct 5, 2026
You announce a change. Someone asks whether their input can change anything.
Inside a Change Announcement practice: a skeptical question, a concrete answer, and coaching that helps separate settled facts from open decisions.
Read articleproduct design · Oct 5, 2026
You are worried about a friend. Can you say it clearly and leave room for an answer?
A Conversations demonstration about naming a concern: the learner's words, a fictional friend's response, and feedback for another practice.
Read articleproduct design · Oct 5, 2026
You asked for a raise. Your manager offers a staged increase. What next?
Step inside a Salary Negotiation practice: the exchange, the Coach's next move, and the feedback you can use in your next conversation.
Read articleresponsible technology · Oct 3, 2026
Feedback Tied To The Moment: How Conversations Reviews A Practice
How Conversations answers the question every practice ends with: feedback at three levels, each tied to a moment in the room.
Read articleresponsible technology · Sep 11, 2026
The product layer of AI safety
Recent frontier safety reports cover model capability and misuse. Most people meet AI inside a product, and that layer has its own rules. Here are ours.
Read articleproduct design · Sep 5, 2026
We built a room, not a chatbot
Why realistic AI conversation practice needs distinct fictional participants, ordered turns, durable state, useful feedback, and human-controlled boundaries.
Read articleproduct design · Sep 1, 2026
Beyond better prompts: the missing operating system for LLM products
Why capable LLM products need durable state, orchestration, evaluation, boundaries, and recovery around the model, not only better prompts.
Read articleresponsible technology · Sep 1, 2026
Guard, don't gag
Why responsible boundaries work best when they are specific to the action, proportional to the risk, and quiet during ordinary use.
Read articleresponsible technology · Sep 1, 2026
Memory should track meaning, not just messages
Why longer histories do not automatically create useful continuity, and what a returning experience actually needs to preserve.
Read articleresponsible technology · Sep 1, 2026
The glass kitchen: transparency by architecture
Why useful explanations have to be designed into an AI product rather than reconstructed after something goes wrong.
Read articleproduct design · Sep 1, 2026
What we mean by a Cognitive OS
A practical definition of the product architecture around an LLM: durable state, orchestration, evaluation, boundaries, and recovery.
Read articleproduct design · Aug 31, 2026
An editorial room, not a conference call
Why multiple AI perspectives need roles, sequence, and synthesis rather than a louder pile of simultaneous opinions.
Read articleresponsible technology · Aug 31, 2026
Anticipation without manipulation
How a product can prepare useful next steps without pretending to predict a person or quietly steering their choices.
Read articleresponsible technology · Aug 31, 2026
Attention without surveillance
Why a product should adapt from explicit, bounded signals instead of treating exhaustive observation as the price of personalization.
Read articleproduct design · Aug 31, 2026
Complete learning materials, not a pile of fragments
Why generating each worksheet, lesson, scaffold, and assessment separately shifts the integration burden back onto teachers.
Read articleproduct design · Aug 31, 2026
Consistency beyond persona prompts
Why a product voice needs observable behavior, boundaries, and variation rather than a long list of personality adjectives.
Read articleresponsible technology · Aug 31, 2026
Continuity without turning attention into surveillance
Why a returning experience should remember what matters while keeping collection, inference, and control within clear boundaries.
Read articleresponsible technology · Aug 31, 2026
Depth without performance
Designing an attentive interaction without simulating intimacy, overclaiming understanding, or filling every pause.
Read articleproduct design · Aug 31, 2026
Designing for participation, not just answers
A practical design test for learning and practice products: what useful work does the person still get to do?
Read articlecompany · Aug 31, 2026
From applications to shared capability
A company perspective on building distinct products without rebuilding every hard-won capability from the beginning.
Read articleresponsible technology · Aug 31, 2026
Make positions inspectable
Why a useful AI collaborator should be able to show what it currently believes, why it believes it, and what could change its mind.
Read articlecompany · Aug 31, 2026
One family does not mean one product repeated five times
How ForeverLearning AI uses shared product principles without flattening conversation, learning, reflection, and authored worlds.
Read article