> For the complete documentation index, see [llms.txt](https://studio-docs.sandbox.game/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://studio-docs.sandbox.game/working-with-ai-agents/how-to-think-about-ai-assisted-development.md). # How to Think About AI-Assisted Development ## Opening: The Problem or Principle AI-assisted development works best when you treat the AI as a capable collaborator, not an automatic builder. The quality of the result depends on the context you provide, the size of the task, and how carefully you review what comes back. In Sandbox Studio, the strongest results come from short cycles: explain the goal, give context, ask for a plan, build in small parts, then review the result before moving on. ## The Core Explanation ### AI Helps You Think, Not Skip Thinking AI can help you plan features, explain engine concepts, write code, find likely bugs, and suggest next steps. It cannot fully understand your intent unless you provide it clearly. A weak prompt asks the AI to “make a game mechanic” without explaining what the player should experience. A stronger prompt says what the mechanic should do, where it should happen, what files or systems matter, and what should stay unchanged. Use AI to reduce the blank page, explore options, and handle repetitive work. Keep ownership of the creative direction, final behavior, and quality bar. ### Context Is the Working Surface The AI works from the information available in the chat. If important details are missing, it may make reasonable guesses that do not match your project. Useful context includes the feature goal, current behavior, desired behavior, relevant files, known errors, engine constraints, and anything you already tried. For Sandbox Studio, this can include scene setup, node names, node behavior, asset paths, or the player interaction you want to create. {% hint style="info" %} *NOTE: More context is not always better. Give the AI the information that affects the task. Extra unrelated details can pull the answer away from the real problem.* {% endhint %} ### Smaller Tasks Produce Better Results AI performs better when the work is broken into focused pieces. Asking it to design, build, debug, polish, and document a full feature in one request creates too many chances for drift. A better pattern is to move feature by feature. Start with the goal. Ask for a plan. Approve the approach. Build the first piece. Test the behavior. Then continue. This keeps the work understandable. It also makes mistakes easier to catch while they are still small. ### Review Is Part of the Workflow AI can sound confident when it is wrong. Treat every answer as a draft until you inspect it. Review the result against your original goal. Check whether the AI changed files outside the requested scope. Look for invented APIs, missing setup steps, unclear naming, or behavior that only works in the explanation. For code changes, run the supported checks for the project and test the feature in the editor when possible. ### Redirect Early When the AI Drifts When the AI starts solving the wrong problem, stop and restate the goal. Do not keep adding corrections to a confused direction for too long. Good redirection sounds like this: “Pause. The goal is not to redesign the whole system. Keep the existing player controller and only add the interaction prompt.” If the chat becomes overloaded with old decisions, start a new chat and bring forward the current goal, the relevant files, and the decisions that still matter. ## Practical Guidance * Start with the outcome you want the player or user to experience. * Give the AI the relevant project context before asking it to build. * Ask for a plan before large or unclear changes. * Work in small feature slices instead of one large request. * Review every AI answer before accepting it. * Redirect the AI as soon as it moves away from the goal. * Start a new chat when the conversation carries too much outdated context. * Keep final judgment with the creator, not the model. ## Common Mistakes | Asking for too much at once | Break the work into planning, implementation, debugging, and polish. | | ----------------------------------------------- | -------------------------------------------------------------------------------- | | Giving a vague prompt | Explain the desired behavior, the current state, and the constraints. | | Accepting AI output without review | Check the result against the goal, project rules, and actual editor behavior. | | Continuing after the AI misunderstands the task | Stop, restate the goal, and narrow the scope. | | Treating AI as the source of truth | Use AI as a collaborator. Trust the project, the engine, and your tested result. | ## The Takeaway AI-assisted development is a faster way to think, build, and revise when you stay in control of the direction. Give clear context, work in small cycles, and review the result before moving forward. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://studio-docs.sandbox.game/working-with-ai-agents/how-to-think-about-ai-assisted-development.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.