Hi, everyone!
Welcome to the new Amazon Developer News & Events! This will be a weekly (or occasionally bi-weekly) roundup of anything interesting happening in the world of Amazon Developer. That includes Alexa, Fire TV, Ring, and anything else happening in our orbit.
As for me, I’m Moses Roth, a Developer Advocate at Amazon. You can connect with me here. Feel free to say hi, always happy to chat about anything in the dev world (or movies or books).
Now for the news!
It was a big week for Fire TV: we launched the Amazon Devices Builder Tools for AI-Powered Development, which includes a Model Context Protocol (MCP) Server and Agent Skills for FOS and Vega. Check out my colleague Pavan Bhat’s guide to that.
We also launched Fire OS 14, a major upgrade from FOS 8, the previous version. Check out our guide on developing for FOS 14 to learn more.
And my colleague Senior Dev Advocate Anisha Malde published a guide to sharing Vega code across Android TV and Apple TV.
We have some other AI and AWS updates and posts for you too, so let’s dive in!
Dev Updates
Updated retry behavior for AWS SDKs and Tools — New retry defaults (faster transient retries, a 1,000 ms throttling base delay, and a retry quota even for legacy-mode SDKs) become default in November 2026; opt in now with AWS_NEW_RETRIES_2026=trueand test in non-production, since apps fail faster on sustained outages and can surface errors sooner.
Announcements
Amazon Devices Builder Tools for AI (MCP Server + Agent Skills)
- A standalone npm package that adds an MCP Server, Agent Skills, and steering docs to your AI coding agent so it has Amazon Devices context for setup, IAP and feature integration, performance analysis, crash debugging, and testing; install with
npx -y @amazon-devices/amazon-devices-buildertools-mcp@latest init-context(Node.js 18+, Vega SDK v0.22+), and it works with Cursor, Cline, Kiro, Claude Code, GitHub Copilot, and Amazon Q.
Developing for Amazon Fire TV devices running Fire OS 14
- A new developer guide for Fire TV on Fire OS 14 covers Android 14 down through 12 (API levels 34–31). It’s the doc to skim before your next release so you target the right levels across the device fleet.
Amazon Bedrock expands request-level usage attribution
- You can now tag any
InvokeModel/InvokeModelWithResponseStreamcall with metadata like team, project, or environment across the wholebedrock-runtimeendpoint (previously Converse-only), giving multi-team deployments proper chargeback visibility.
Introducing the Claude Platform on AWS
- Now generally available: the full native Claude API surface (Managed Agents, code execution, skills, web search/fetch, Files API, MCP connector, prompt caching, citations, batch) through AWS IAM, CloudTrail, and billing, with day-one feature parity — distinct from Bedrock in that Anthropic operates the service and processes data outside the AWS boundary.
Agent view in Claude Code
- A unified CLI interface for managing multiple parallel sessions: run
claude agentsto see which need input, which are working, and which are done, and background jobs with/bgorclaude --bg [task], now in Research Preview.
Dev Posts
Two first-party reads lead this week. Sharing your Vega app across platforms (Android TV, Apple TV & Web) is where I’d start if you’re weighing a multi-platform port, because it puts a real number on how much of your React Native code carries over. Then Accelerate Fire TV app development using your preferred AI coding agent is the practical follow-up: how to give whichever agent you already use the Fire TV context it’s missing.
Sharing your Vega app across platforms (Android TV, Apple TV & Web)
By Anisha Malde
You can reuse up to 85% of your Vega React Native code across Fire TV, Android TV, Apple TV, and the web. The post walks through how the shared codebase works, which platform-specific pieces remain, and what the trade-offs are. If you’re building for Fire TV and weighing whether to expand to other TV platforms or the web, this is the scoping read to start with.
Accelerate Fire TV app development using your preferred AI coding agent
By Pavan Bhat
Amazon Devices Builder Tools for AI gives your coding agent, whichever one you use, the right context for building, debugging, and migrating Fire TV apps. The post shows how to wire it into your workflow so the agent stops hallucinating Fire TV APIs and writes code that actually runs on the device. Worth setting up if you’re already using Claude Code, Cursor, Copilot, or similar for Fire TV.
Improving quality and robustness in LLM-based text-to-speech systems
By Ammar Abbas
Amazon researchers shared three techniques that meaningfully improve LLM-based TTS. Low-rank adaptation (LoRA) with locale-specific data lets a cloned English voice speak French, German, or Spanish naturally without losing speaker identity. Classifier-free guidance generates synthetic reference samples and lifts expressiveness 5–20% across nine locales. And a stack of chain-of-thought reasoning, guardrails, agentic regeneration, and smarter data filtering knocks critical errors down to under one second per hour on long-form text. For Alexa and any device team shipping voice, this is the path to more natural, multilingual, production-reliable speech.
Test Driven Development (TDD) with Kiro: this is how it should feel
By Mike George
Kiro now supports automating TDD through its hooks feature, which fires actions on IDE events like file saves. Set a “Pre Tool Use” hook on the write tool that enforces red-green-refactor, and Kiro writes a failing test first, then minimal code to pass, then refactors, every time, no skipping. The post walks the setup on a toy Monty Hall example and a more realistic REST API built with Kiro’s spec-driven workflow. The argument: keep TDD’s quality benefits without the discipline tax by baking the rules into a hook so the agent can’t cut corners.
How Amazon reduced fulfillment center verification time by 60% with Amazon Nova
Amazon’s fulfillment centers run on precision. Every module has to be verified before a facility goes live, a process that once took 2,200 people inspecting 1,300+ modules over 2,000 hours per deployment cycle. Data & AI Program Manager Akhil Anand explains how Amazon Global Engineering Services partnered with AWS to build IORA (Internet Operations Readiness Assistance) on Amazon Nova Pro: workers photograph equipment instead of manually checking it, and IORA identifies each item and flags issues. The result is a 60% cut in verification time and a concrete case study for Nova Pro on real-world image-recognition workloads.
AI is the biggest technology transformation in our lifetime, says Amazon CEO Andy Jassy
Amazon CEO Andy Jassy joins CNBC’s Jim Cramer on Mad Money to talk through the quarter and where AI is headed. He calls AI the biggest technology transformation of our lifetime and gets into how it’s changing Amazon’s supply chain and the products it ships.
The race to make robotaxis mainstream (w/ Zoox CEO Aicha Evans)
Zoox CEO Aicha Evans joins Rapid Response on Zoox’s strategy as an Amazon subsidiary: where robotaxis stand today, the competition, and what it takes to get riders to trust a car with no steering wheel.
Feedback
Questions or comments? Have an item for next week’s post? Let me know!