Saturday guide: the week of systems that can't stop

This week brought together very different cases that share the same underlying problem: what to do while something slow runs underneath, whether it's an AI call, a database migration, or a compression bug hidden inside a library used by half the world. Set aside time for today's tutorial before moving on to the reads: they fit directly into the same conversation.

To do: this week's tutorial

[ADK and Gemini 3.8 Live show how to avoid silence in voice agents](/inteligencia-artificial/adk-e-gemini-3-8-live-mostram-como-evitar-silencio-em-agentes-de-voz), by Alan Andrade.

It's the kind of fix that only shows up when the agent freezes live, not in a quiet local test.

To read: three from the week

• [OpenAI resumes frontier model training and reveals a Brotli bug used by Node.js, Bun, and Envoy](/noticia/openai-retoma-treinamento-de-modelos-frontier-e-revela-falha-no-brotli-usada-por): The pause OpenAI put on itself ended up exposing a bug that can affect any service that uses Brotli for compression, as is the case with Node.js, Bun, and Envoy. (Redação iMasters)
• [How to do a major PostgreSQL upgrade without downtime in a distributed cluster with Spock](/banco-de-dados/como-fazer-upgrade-major-do-postgresql-sem-downtime-em-cluster-distribuido-com-s): Hadi documented, in practice, what happens between triggering the upgrade and the moment the cluster actually stops complaining. (Roberto Diniz)
• [Kimi K3.1 Pro arrives at Vercel's AI Gateway with 1M token context and open weights on the way](/front-end/kimi-k3-1-pro-chega-ao-ai-gateway-da-vercel-com-contexto-de-1m-de-tokens-e-pesos): With 1 million tokens of context already in Vercel's AI Gateway, Moonshot AI's model enters the race even before opening its weights. (Carina Ferreira)

To answer: this week's conversation

On the iMasters Forum: Validating keys in i18n dictionaries with TypeScript for React projects, started by Patricia Gomes, with 4 replies so far.

The weekend is good for that: tell us on the forum what you're working on.