Weekend Showcase Challenge: FutureNews AI
Tomorrow's headlines, written before they happen.
Vision and What It Does
FutureNews AI is an autonomous AI newspaper that writes fictional news from the future.
Unlike a normal prompt-in, article-out generator, FutureNews AI remembers what has already happened. Each new edition uses previous editions, the current world state, and ongoing storylines to decide what should happen next.
For example, one edition might introduce an AI-managed city. The next could introduce AI-powered transport, followed by security concerns and eventually global regulation
The goal is to create an evolving fictional timeline where every event has consequences.
Visitors can browse existing editions without triggering AI generation.
Reading the future should never create the future.
How I Built It
The backend runs on AWS Lambda behind Amazon API Gateway.
One of my first bugs was that /latest and /editions were accidentally reaching the same generation logic as /generate . This meant simply loading the website could create a new edition.
I fixed this with explicit route dispatch, separating read-only operations from AI generation.
Other challenges included DynamoDB Decimal serialization, the /prod API Gateway stage prefix, and configuring the correct global inference profile for Claude Sonnet 4.5 in Amazon Bedrock.
For this challenge, I also added an Agent Self-Review card. After generating an edition, users can see the agent's review score, approval result, revision status, and timeline validation.
AWS Services and Architecture
FutureNews AI uses:
- Amazon EventBridge Scheduler — triggers the autonomous agent.
- AWS Lambda — runs the API and generation pipeline.
- Amazon Bedrock — generates events, reviews drafts, and writes articles.
- Amazon DynamoDB — stores editions and persistent world memory.
- Amazon API Gateway — provides the public API.
- AWS Amplify — hosts the React frontend.
- Amazon CloudWatch — stores application logs.
- AWS IAM — manages permissions.
The architecture is:
EventBridge Scheduler → Lambda → DynamoDB world memory → Amazon Bedrock → Self-review + timeline validation → Save new edition and world state
The frontend only uses read-only routes, keeping browsing separate from AI generation.
What I Learned
The biggest lesson from building FutureNews AI was that persistent memory changes how an AI application behaves.
Instead of generating unrelated content every time, the agent can build on previous events and create continuity between stories.
I also learned the importance of separating read-only and side-effecting operations. A visitor browsing the website should not accidentally trigger an AI generation request.
Finally, working with Amazon Bedrock taught me more about inference profiles and AI workflows that include generation, validation, and self-review.
Project Screenshots
Screenshots of FutureNews AI, including the homepage, generated editions, and Agent Self-Review:
Links
FutureNews AI explores one simple idea: what if an AI-generated future could remember its own past?
Instead of creating isolated headlines, it builds a fictional world where yesterday's events influence tomorrow's news.
Loading article