AWS Builder Center

Weekend Showcase Challenge: Drift

Most creative AI tools begin with an empty box. They wait for someone to arrive, invent a prompt, and ask for an output. Drift reverses that relationship. It is an autonomous daily micro-magazine that creates a small, finished creative object before I wake up: an original short poem, a four-color story, and one offline invitation to notice or make something.

Weekend Showcase Challenge: Drift — A Daily Field Note That Learns in Public

#application #challenge

The vision: creativity without the empty prompt

Most creative AI tools begin with an empty box. They wait for someone to arrive, invent a prompt, and ask for an output. Drift reverses that relationship. It is an autonomous daily micro-magazine that creates a small, finished creative object before I wake up: an original short poem, a four-color story, and one offline invitation to notice or make something.
Every issue is shaped by the current weather in New Delhi, the rhythm of the day, and lightweight reactions to recent issues. A reader never has to become a prompt engineer. They can simply choose “keep this direction,” “surprise me more,” “make it calmer,” or “push it bolder.” Drift carries those bounded signals into the next day’s editorial brief.
The goal is not to generate more disposable content. It is to create a daily creative ritual with continuity. Weather becomes material rather than a forecast, while yesterday’s reactions become compact editorial memory rather than an endless chat history.

What changed for the Summer Builds Showcase

I first built Drift as my autonomous-agent project in the Summer Build Series. For the finale, I wanted to improve the part of the idea I care about most: visible adaptation.
The original version already published an “adaptation receipt” with every issue. It explained the high-level editorial shift—for example, “quieter cadence, one vivid contrast, and more concrete imagery”—without exposing private model reasoning. For this showcase version, I added a learning ledger. It visualizes the four accumulated reader signals that shaped the next editorial brief, shows the strongest current direction, and pairs that memory with the archive of generated issues.
That finishing touch makes Drift’s claim testable. A reader can see the input signal, inspect the accumulated preference, compare the published style shift, and then browse earlier issues. Drift does not pretend that this is model training. It is deliberately small, bounded editorial memory, made public.

How I built it

I began with a strict definition of a successful daily artifact. Every Drift issue must contain a headline, subtitle, four-to-eight-line poem, exactly four named colors with valid hex values, a creative prompt, a weather note, and a concise adaptation note. That contract became a structured prompt for Amazon Nova 2 Lite through the Amazon Bedrock Converse API.
At 5:30 each morning, Amazon EventBridge Scheduler invokes an AWS Lambda function. The function checks Amazon DynamoDB for an issue with the current date and exits if one already exists. This makes retries safe and prevents duplicate model calls. If the date is new, Lambda reads the seven most recent issues, aggregates their reaction counts, and retrieves the current local weather. A temporary weather failure does not stop the run; Drift records that the context was unavailable and continues.
Lambda turns those inputs into a compact editorial brief. More “calmer” than “bolder” reactions creates guidance for a gentler cadence and softer transitions. “Surprise me more” asks the model to avoid the obvious weather metaphor. Recent titles become a do-not-repeat list.
Nova returns structured JSON, but the output is never written directly to storage. The Lambda function extracts and parses the response, validates poem length, checks every hexadecimal color, limits field sizes, and rejects incomplete editions. Only a valid issue reaches DynamoDB, using a conditional write that allows one edition per date.
The interface is designed like a collectible field note rather than an administration dashboard. Oversized editorial typography, a weather dial, daily swatches, the adaptation receipt, the new learning ledger, feedback controls, and the archive all support one idea: autonomy should be visible, legible, and enjoyable.

AWS services and architecture

Drift uses:
  • Amazon EventBridge Scheduler for the timezone-aware daily trigger and retry policy.
  • AWS Lambda to gather context, construct the brief, invoke the model, validate the result, and serve API requests.
  • Amazon Bedrock with Amazon Nova 2 Lite to generate each structured creative issue.
  • Amazon DynamoDB to store immutable issues and compact reaction memory with on-demand capacity.
  • Amazon API Gateway HTTP API to expose read and feedback endpoints to the website.
  • AWS SAM and AWS CloudFormation to keep the backend reproducible as infrastructure as code.
The autonomous flow is:
 EventBridge Scheduler → Lambda → recent DynamoDB memory + current weather → Amazon Bedrock/Nova → validation → conditional DynamoDB write 
The reader flow is:
 Drift website → API Gateway → Lambda → DynamoDB 
Feedback travels through the same API and atomically updates an issue. The generation route remains private, so public visitors cannot trigger unexpected Bedrock usage. The system performs one planned model request per day, while ordinary visits read work that has already been generated. This keeps the project serverless, responsive, and cost-aware.

What I learned across the summer

The summer challenges changed how I think about AI products. First, I learned to begin with a crisp human need instead of a model capability. Drift is memorable because “a poem, palette, and prompt before breakfast” is a promise people can understand.
Second, I learned that an agent becomes compelling when it has a ritual, not just a trigger. Scheduling proves that Drift can run without me, but continuity gives the automation meaning.
Third, I learned that memory does not need to be enormous. Four carefully chosen signals can create a strong sense of evolution when the interface shows what changed. The new learning ledger made that lesson concrete: observability can be part of the product experience, not only an engineering dashboard.
Finally, I learned to treat model output as untrusted application input. Structured prompting helps, but validation, length limits, fallbacks, idempotency, conditional writes, and controlled permissions are what turn a creative prototype into a dependable agent.

Try Drift

A builder who inspired me

I’m tagging @Lewis Sawe, whose project The Museum That Grows showed how an always-on agent can create continuity, world-building, and anticipation one artifact at a time. That idea—that a scheduled agent can feel like a living creative practice—inspired me to make Drift’s archive and learning process more visible.
Drift began as an experiment in creative autonomy. It ends the summer as something closer to a public practice: weather in, reactions remembered, and something new out every day.
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