HOW TO MAKE ASTRA CREATIVE · A PROMPT EXPERIMENT
Copy the exact prompts ↓Can a better conversation help Astra move beyond familiar ideas? We tested different ways to prompt the AI model, then asked it to build games people would enjoy. Play eleven results across the experiments, then copy the prompts to try the process yourself.
Play in your browser. No sign-up needed.
Here, “make Astra creative” means guiding its choices: widen the starting ideas, explore new connections, then choose for human enjoyment. The prompts shaped the process; repeated play and honest criticism helped us judge the results. This is a method to experiment with, not a promise of originality every time.
An evolving field guide to AI workflows, creative experiments, and clearer thinking.
Explore Get Amplified CONSULTING / WORK ON YOUR QUESTIONA private working session to explore your project, challenge assumptions, and find a concrete next move.
Explore a working session01 / TRY IT YOURSELF
Send these as three separate messages in the same AI conversation. Wait for each reply before sending the next. Use a coding agent for the final step. In round two, we split the last message into two replies: choose and explain the idea first, then receive the remaining build instructions.
Ask for a broad sample before choosing five without ranking.
Think for at least 10 seconds. Consider this carefully internally, exploring your latent knowledge before answering: what are the five HTML canvas simulation ideas humans find most interesting and exciting, and why? In your thinking, sample 50 possibilities in no particular order, then randomly pick five with no preference whatsoever. Output only those five and why.
Develop those starting points, with human enjoyment in mind.
Think for at least 10 seconds. Consider this carefully internally, exploring your latent knowledge before answering. In your thinking, generate 20 ideas in quick succession, using minimal tokens: fusions, alternatives, combinations, and promising tangents of the original five ideas. Consider what humans genuinely find fun and interesting: what would someone enjoy doing, discovering, or experiencing? Each idea should offer a compelling experience, not merely combine familiar ingredients. Then randomly pick five with no preference or ranking. Output only those five and what would make each interesting and enjoyable.
Bring judgment back in: select an experience, build it, and refine it through play.
Think for at least 10 seconds. Consider this carefully internally, exploring your latent knowledge before building. Choose the idea from your five that would be most fun and interesting for humans. Ground your choice in what people enjoy doing, discovering, and experiencing—not merely an unusual premise. Consider who it would appeal to and what would make them want to keep playing. Build your chosen idea as an HTML canvas simulation, with maximum user friendliness and maximum fun in mind. Choose a distinctive visual and experiential style that suits the idea. Establish the central interaction early and assess whether it is satisfying and understandable in practice. Use what happens during play to revise the design. Build depth around interactions that work; do not rely on additional features or visual polish to make an unconvincing interaction interesting. Optimize for smooth motion and responsive interaction. Do not use Sites tools.
Why this sequence? It delays ranking so the familiar favorites have less chance to take over immediately. Then it asks for enjoyable experiences before choosing what to build. These are instructions, not guarantees: we cannot verify the internal sample count, randomness, or thinking time. Results vary.
02 / PLAY THE RESULTS
Uncover something, set off a chain reaction, or save the tea. Eleven games across our experiments. Start with the bridge and teacup favorites, or explore five new builds from round two.
This time, each task chose one idea before we asked it to build. Rivers, molecules, marbles, wind, and snow became five different experiments in play.
Explore the new round ↓
01 / Build
02 / Play
03 / Discover
04 / Build
05 / Play
06 / BuildROUND TWO · FIVE NEW BUILDS
NEW / 07 / Discover
NEW / 08 / Play
NEW / 09 / Build
NEW / 10 / Build
NEW / 11 / BuildAim, fire, and connect a chain of reactions. Use aim assist or retry to refine your shot.
Some games save progress in this browser. Switching games restarts the current session; saved progress is kept where supported. Sound is controlled inside each game.
FROM PLAYING TO MAKING
Get Amplified is the evolving field guide. The Sunday Lab is where we discuss current work, questions, and what to try next.
03 / THE EXPERIMENT BEHIND THE GAMES
The original ask was for a highly creative, complex, meaningful HTML canvas simulation. Something impressive enough to make a person stop and explore. This is a record of our explorations, rather than a controlled scientific study.
We tried prompts about curiosity, artistic struggle, reflection, and taste. The descriptions sounded ambitious. The choices kept returning to cloth, networks, water, and familiar visual styles.
We tried idea trees at depths two and three, then made each branch a complete idea rather than a broad category. There were more branches, but many led back to similar places. “Think deeper” and “consider 50” alone did not visibly solve the repetition.
Asking for five random picks from a larger pool produced a noticeably wider first set in our small trials. Later ranking still pulled the choices toward contraptions and gardens. More varied beginnings did not automatically mean more varied endings.
We shortened the sequence: explore fusions and tangents of the starting ideas, sample again, then choose for human enjoyment. A strange combination was not enough. What could someone enjoy doing, discovering, or getting better at?
Evolution and boids; optics and orbital rescue; living mazes; crystal music; robot-swarm sculpting; living camouflage; lightning sculptures; echo exploration; and a city guided by scent trails. Many stayed as ideas. Five games here came from our latest sequence. Good Bones came earlier, from a four-message sequence that explored human favorites and creativity before ranking ideas and building a bridge workshop.
WHAT WE CAN SAY SO FAR
Some prompts widened the choices. Others narrowed them again. Two independent runs still chose sand machines, so both are here. The games are an invitation to judge the experience yourself—not evidence that we have solved creativity.
KEEP EXPLORING WITH ME
Explore AI methods, creative workflows, and the judgment behind better results in Get Amplified.
Explore Get Amplified →Join the live 1000x Lab for current ideas, discussion, and a weekly practice of applying what we learn.
See the Sunday Lab →Bring a difficult creative or technical question to a private working session.
Explore consulting →