I tried Kiro.

These days I was in the mood for a “cook-and-serve” project: a minimal AWS soffritto (just enough infrastructure to make it real), a pinch of AI (apparently mandatory in 2025), and no more than ~24 hours (not consecutive) to build something end-to-end inside Kiro.

After the last event I organized, I got an idea: a Telegram bot that, every day, publishes into a dedicated channel the items from a list of RSS feeds-only those from the last 24 hours—then translates and summarizes them in Italian.

What I shipped:

Two branches, two stories

In the repo you’ll find two branches:

  • master: the final version, driven by me — working and (overall) satisfying
  • kiro-managed: what Kiro produced starting from my initial prompt (included in the documentation), up to a working project

The part I found most interesting is not “how much code Kiro wrote”, but how it structured the work: it turned the prompt into a plan and kept asking follow-up questions (for example: MVP-first vs a more complete build). The generated spec is here:

https://github.com/palumbou/another_rss_telegram_bot/tree/kiro-managed/.kiro/specs/another-rss-telegram-bot

Architecture (AWS serverless)

The bot is designed to be generic and reusable, and it runs on AWS with a fairly linear serverless architecture:

AWS components

  • EventBridge Scheduler: daily trigger
  • AWS Lambda (Python 3.12): orchestration (fetch RSS → filter last 24h → dedup → summarize → post to Telegram)
  • DynamoDB: deduplication storage with a 90-day TTL
  • Secrets Manager: secure Telegram token storage
  • Amazon Bedrock: Italian summary generation (default: Amazon Nova Micro, eu.amazon.nova-micro-v1:0)
  • SQS Dead Letter Queue (DLQ): error handling and retry path
  • CloudWatch: logs, metrics, and a monitoring dashboard
  • CodePipeline + S3: CI/CD and artifact storage

Code components (high level)

  • src/lambda_handler.py: main entry point and orchestration
  • src/rss.py: RSS management (feed parsing + normalization)
  • src/telegram.py: Telegram Bot API integration
  • src/summarize.py: Bedrock summarization + fallback logic
  • src/dedup.py: DynamoDB-based deduplication
  • src/config.py: configuration and environment management

If you want the full infrastructure notes, the repo contains an infrastructure guide and a single CloudFormation stack that includes Lambda, DynamoDB, EventBridge, CodePipeline/CodeBuild, and related resources.

Kiro in practice (no fanboying)

There are many “AI-first” IDEs built on top of (or around) the VS Code ecosystem, with different models, pricing, and the usual mix of great ideas and sharp edges. I personally ran into issues like this more than once:

https://github.com/kirodotdev/Kiro/issues/3897

Kiro (AWS) currently offers new users a 30-day trial with 500 credits. I tested it exactly like that: I let Kiro pick the model at first, then I switched to Claude once I took back the steering wheel and started making the key decisions myself.

Conclusion

Kiro is a tool. Try it.