--- title: "From a Brief to a Working Demo with Astra" description: "How we made three original AI-assisted browser experiments: visual direction, structured data, working controls, honest examples and reusable build briefs." author: Power of AI date: 2026-09-10 url: https://powerofai.ca/workflows/from-brief-to-working-demo --- # From a Brief to a Working Demo with Astra How we made three original AI-assisted browser experiments: visual direction, structured data, working controls, honest examples and reusable build briefs. Build notes / Behind the Possibility Studio We wanted this site to do more than tell people that AI can help them make things. The Possibility Studio is our answer: an original architectural miniature, an evidence viewer and an agent rehearsal. Astra in Codex helped design and implement the software. An image-generation tool made the observatory concept. The browser exhibits themselves use ordinary local code. ## Give the project a job and a feeling The creative brief was to make curiosity tangible. We kept the blue identity, chose a miniature observatory as the visual centre and designed three experiments around different kinds of work. One invites you to build a place. One lets you trace a claim. One reveals the stages of an agent workflow. Each needs a useful interaction within a minute of arriving. Aesthetic direction belongs in the brief alongside function. Describe the scale, palette, visual hierarchy and feeling you want. Our concept asks for a small architectural world with pale stone, cobalt terraces, gardens and a bridge. The interface gives that world room to breathe. On smaller screens, the controls move below the scene so the exhibit remains usable without shrinking the labels into decoration. ## Make the underlying structure small enough to explain The world is a collection of three-dimensional shapes with positions, materials and build stages. Its renderer batches repeated shapes and redraws when you change the view or settings. The evidence viewer is a collection of claims, source passages and authored relationships. The workbench is a brief plus three stages whose results follow explicit rules. None requires an ongoing model call to respond to a visitor. AI assistance helped create the program that now handles those interactions. This is a useful distinction when planning your own project. Some features need model inference at use time; others need well-designed software that a model can help you write. Start with the simpler structure when it meets the goal. It gives you something repeatable to test and a clear place to investigate when a control or an exported result behaves incorrectly. ## Ask for a reviewable first version, then use it A strong first request includes a visible output, a small set of controls and a way to keep the result. Ask the assistant to build a local preview and explain the important choices. Use the preview before adding features. Rotate the world, remove a source, change the brief after a review. Those actions tell you more about the experience than a screenshot of its default state. Then give concrete feedback. A mobile control is too narrow. An excluded source still sounds authoritative. A downloaded report includes a stage that never ran. Each observation points to a behaviour that can be corrected and checked again. Visual polish and dependable interactions reinforce one another when the feedback describes what a visitor is actually trying to do. ### A request you can adapt Help me build one original local browser experiment. First define its purpose, visual direction, data model, controls and export. Implement a small complete version. Show me desktop and mobile previews. Check changed inputs, empty states, reset and downloaded output. Explain what is simulated and what actually runs. Keep publishing as a separate decision. ## Keep the build story attached to the finished thing The observatory illustration and the working 3D miniature are deliberately labelled as different artifacts. The Night Library records are fictional. The evidence relationships and agent responses are authored teaching cases. Those labels help readers understand what the exhibit demonstrates: an interface and a workflow they can adapt, rather than a hidden performance claim about an AI model. Keep your own build notes just as concrete. Record the original request, the major design choices, the controls that work and the limitations that remain. Export a small artifact a reader can inspect. A tutorial becomes more useful when it shows the path from an idea to specific working behaviour, and leaves enough room for someone else to make a version with a different purpose. ## Take this into your next task Start with a small complete experience. Use AI to help create it, then judge the actual controls, evidence and output. [Build a project brief](/showcase#brief-builder). ## Questions people ask ### Is this a live AI demo? No. Astra in Codex helped create these original browser programs. Their controls run locally with authored rules; they do not call a model. ### Can I keep what I make? Yes. Download the current output, or explicitly save a world to your local creations shelf with its settings and camera view. The notebook lets you reopen worlds and export or import a library backup. Unsaved experiments last only while the page is open. ## Sources and editorial notes Reviewed 2026-09-10. Product capabilities depend on the app, plan, region and workspace settings. Workflows and prompts are authored teaching material, not recorded model results or measured time-saving claims. ## Keep going - [Build a Little World with AI](/studio/world-builder) - [Make an Evidence Atlas with AI](/studio/evidence-atlas) - [Build an Agent Workbench You Can Inspect](/studio/agent-workbench)