AWS Builder Center

Weekend Creative Agent Challenge: Muse Before You Open

Muse Before You Open is an always-on AWS agent that creates a fresh 15-minute creative brief with GPT-5 mini every day before you ask for it.

Software Engineering Student | Cloud & AI Enthusiast | Learning AWS

Weekend Creative Agent Challenge: Muse Before You Open

Tags: #agents

Vision And What It Does

Creative work often gets blocked before the first mark, word, or note is made. The obstacle is not always a lack of ability; it is the small but real cost of deciding what to make. I built Muse Before You Open to remove that decision. It is an always-on creative agent that produces one practical creative briefing every day, before I open the app or ask it to generate anything.
Each briefing is designed to create momentum in about fifteen minutes. It includes a theme, a focused prompt, a three-color palette, and one constraint. For example, a daily brief might suggest a small scene, a micro-poem, or an object that implies a larger world, then limit the work to three marks or three sentences. This gives a creator enough direction to begin immediately while keeping the result open to personal interpretation.
The creative output is made autonomously. Amazon EventBridge Scheduler runs at 12:00 UTC every day and invokes an AWS Lambda function without a button click, web request, or manual prompt. Lambda retrieves an OpenAI credential from AWS Secrets Manager, sends the brief request to GPT-5 mini, validates the structured JSON response, and saves the finished briefing to Amazon DynamoDB. A public Lambda Function URL returns the most recently archived briefing, so the output is ready when I return.

How I Built It

I started with a deliberately narrow scope: one daily text briefing instead of a complex chat interface. The Lambda function is written in Python and has two separate paths. The scheduled path generates and stores a new briefing. The public URL path is read-only and returns the latest saved briefing. This distinction matters because opening the live URL never triggers another model request. The schedule, not a visitor, is responsible for producing the creative artifact.
To prevent the agent from feeling repetitive, Lambda scans the recent DynamoDB records and passes the seven most recent themes to GPT-5 mini as exclusions. This is a lightweight form of memory. It does not require a separate vector database, but it makes the output feel more like a continuing creative companion than a random prompt generator.
I requested an exact JSON shape from the model: a theme, prompt, three-color palette, constraint, and a fifteen-minute duration. The function validates the response before storing it. This was a key implementation decision because it keeps every generated record predictable for the public endpoint and for possible future clients.
The main challenge was secure use of an external model API. The API key is not in Git, environment variables, or the deployment script. Instead, Lambda reads the existing  Open_AI_key  secret only at runtime. Its IAM policy grants access only to that named secret and the specific DynamoDB table. I also used AWS CLI and CloudFormation rather than creating resources manually in the console. The PowerShell deployment script packages the local Lambda source into S3 and updates the infrastructure stack repeatably.

AWS Services Used And Architecture Overview

  • Amazon EventBridge Scheduler triggers the agent daily without user interaction.
  • AWS Lambda runs the Python generation and public read-only endpoint logic.
  • AWS Secrets Manager securely supplies the OpenAI API key during execution.
  • Amazon DynamoDB stores daily briefs and provides recent-theme history.
  • AWS Lambda Function URLs provide the live endpoint for the latest briefing.
  • Amazon CloudWatch Logs records scheduled invocations and saved themes for verification.
  • Amazon S3 stores Lambda deployment artifacts uploaded by AWS CLI packaging.
  • AWS IAM supplies least-privilege roles for Lambda and EventBridge Scheduler.
EventBridge Scheduler --\> Lambda --\> OpenAI GPT-5 mini
| |
v v
DynamoDB \<--- validated creative brief
|
v
Lambda Function URL (latest brief)

What I Learned

This build showed me that useful agents do not need to be conversational. Scheduling changes the product experience: instead of opening a tool and requesting inspiration, I return to an idea that is already waiting. I also learned how EventBridge Scheduler invokes Lambda through a dedicated IAM role, how CloudFormation packaging turns local code into S3 deployment artifacts, and how to combine Secrets Manager with an external API without exposing credentials.
The deployment also reinforced the value of evidence. The enabled daily schedule, CloudWatch invocation logs, and DynamoDB record together show that the creative output was generated without manual initiation. My next iteration would add a static visual interface and feedback controls, allowing creators to rate briefs and gradually tune the themes toward their preferred style.

Links

Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article