July 20, 2026: Child-Safe Apps & Strands Agent Harnesses

Hi, everyone!

It’s summer, so we’re pretty focused on events and that means we don’t have a lot of big updates for you while we prepare for our workshop at Chain React next week. We hope to see you there!

There are a couple updates to make sure all apps on the Amazon Appstore are safe for children: the GetUserAgeData API is now returning live responses and there’s a new Parental Consent Revocation Report for COPPA-scoped apps.

On the AI side, Grok 4.3 landed on Amazon Bedrock with configurable reasoning effort and a million-token context window.

AWS Developers posted a 14-part video course on building AI agent harnesses with Strands Agents.

And if you’re looking for some summer reading, I highlighted a couple of blogs that I found super interesting this week.


:hammer_and_wrench: Dev Updates


:loudspeaker: Announcements

:brain: Grok 4.3 is now available on Amazon Bedrock

  • xAI’s debut on Bedrock brings configurable reasoning effort (none through high, set per request), tool calling, structured JSON output, image input, and a 1M-token context window, accessed via the OpenAI-compatible Mantle inference engine with standard IAM auth.

:date: Events

We’re on the program at DevRelCon NYC this Tuesday, and Chain React in Portland the following week has us bringing a speaker and a workshop.

Date Event Location Description
July 22 DevRelCon NYC New York City, US A developer-relations conference for the people who build and run developer communities, where our Developer Advocate Kourtney Meiss is speaking.
July 30–31 Chain React Portland, US Developer Advocates Giovanni Laquidara and Kourtney Meiss are leading a workshop on building an agent harness for your React Native app.
Aug 1–31 Gamescom Cologne, DE Global games industry event with Amazon Games presence.
Aug 4 Ai4 Las Vegas, US Enterprise AI conference; 8,000 attendees across 75+ countries discussing applied AI and agents.
Sep 2 AWS Zurich Zurich, CH AWS community day in Switzerland.
Sep 11 ZurichJS Conf Zurich, CH JavaScript community conference.
Sep 11–14 IBC Show Amsterdam, NL The media, entertainment, and technology show where streaming-device teams gather.
Sep 17–18 Agent Conf Warsaw, PL Developer Advocates Kourtney Meiss and Anisha Malde are giving a talk, and Giovanni Laquidara is leading a workshop.

:speech_balloon: Dev Posts

Short list this week. Two of the three reads are about cost: Five AWS Costs That Surprise People and How to Avoid Them is the practical one, config-level defaults worth checking before your next bill lands, and The cost of saying yes has changed is the more interesting argument: AI made writing code cheap and left the cost of owning it exactly where it was. The third is an Amazon Publisher Services case study on 14X-ing ad revenue with a one-line SDK bridge. And to watch: a 14-part course on building AI agent harnesses with Strands Agents.

:newspaper: Five AWS Costs That Surprise People and How to Avoid Them

By Muhammad Muzammil

The most common bill shocks come from services doing exactly what you told them to. This piece walks through five charges that catch teams off guard and gives practical guardrails for each: the kind of config-level defaults that run up a tab while your attention is on code rather than billing dashboards.

:newspaper: One update to SDKs led to mobile publishers earning 14X more APS revenue from existing inventory

One update to SDKs led to mobile publishers earning 14X more APS revenue from existing inventory
Amazon Publisher Services’ case study on the Mobile SDK bridge from its annual summit, and it opens as a PDF. Activating a bridge takes one or two lines of code, with no change to your existing mediation setup and no new contracts. Four gaming publishers in the beta saw APS revenue from InMobi grow as much as 14X on existing inventory, and one watched flexible media go from 3% to 35% of its APS total inside 28 days.

:film_projector: Building AI Agent Harnesses with Strands Agents

AWS Dev Advocate Morgan Willis leads a 14-part course on building an AI harness, including the tools, hooks, guardrails, and state management that decide whether it holds up in production. Built on the open-source Strands Agents SDK, it covers the agent loop, tools via MCP, hooks and steering, context and memory, multi-agent patterns (agents as tools, graph workflows, swarms), evaluation, and cloud deployment.

:newspaper: The cost of saying yes has changed

By Dalia Abuadas

AI dropped the cost of writing code, but the cost of owning it stayed the same. This GitHub Engineering post proposes treating an AI-generated patch as a probe: hand an agent a constrained attempt (smallest patch, behind a flag, no contract changes, with tests) and use the result to price the real work. The dividing line shifts from “can an agent write this?” to “can a human confidently review and own it?” Auth, privacy, billing, and product contracts still deserve hard scrutiny regardless of diff size.


:open_mailbox_with_raised_flag: Feedback

Questions or comments? Have an item for next week’s post? Let me know!