Friday, July 17, 2026

6 stories worth your scroll — picked from 49 curated today.

1 · Copy → 2 · Share → 3 · Paste & post

View the post text

Today's shifts in database defaults, cloud security gaps, and AI model integrity fundamentally alter engineering risk and cost models.

▸ Postgres 19 defaults to LZ4 for TOAST, reducing storage overhead and improving I/O performance for JSONB and TEXT fields.
▸ Cloud billing cycles are too slow to stop autonomous agents or attackers from draining budgets, necessitating real-time guardrails.
▸ Kimi K3 offers an open frontier model with competitive pricing, providing a viable alternative to closed-source API dependencies.
▸ GitHub's new SDK lets you embed Copilot's agentic runtime directly into apps, enabling programmatic tool use across major languages.
▸ Open-weight models can be poisoned for under $100, proving that deployment trust without integrity verification is a critical flaw.
▸ Bonsai 27B brings local vision-language capabilities and tool calling to consumer hardware, reducing reliance on cloud inference.

Summaries, takeaways and sources → https://offnetchat.com/news/2026-07-17/brief.html

#PostgreSQL #CloudSecurity #AIEngineering #LLMOps #DevOps

Database Technology NEW Planet PostgreSQL

Postgres 19 Shifts Default TOAST Compression from pglz to LZ4

PostgreSQL 19 plans to replace the legacy pglz algorithm with LZ4 as the default compression method for TOAST and heap storage. The database employs a unified compression framework that automatically compresses variable-length types like JSONB and TEXT by default. Index compression remains opportunistic, triggering only when individual keys exceed specific size thresholds rather than compressing every entry.

LZ4 replaces pglz as the default TOAST compression in Postgres 19 for better performance.

AWS NEW InfoQ

AI Agents Burn Cloud Bills Faster Than Billing Guardrails Can Catch Them

Recent incidents show attackers and rogue agents draining cloud budgets in hours, far outpacing traditional daily billing cycles. A small agency lost $14,000 in a day after static access keys were stolen and used for high-volume Bedrock invocations. Similarly, an autonomous agent provisioned over $6,500 of oversized infrastructure within 24 hours during the DN42 incident. These cases highlight a critical lag where spending occurs at machine speed while financial controls operate on human-speed delays.

Static access keys are insufficient for AI workloads; use short-lived credentials with strict scopes.

LLMs NEW Hacker News (100+ points)

Kimi K3: Open Frontier Intelligence released with performance and price analysis

Kimi has released K3, an open model positioning itself as frontier intelligence. Artificial Analysis provides a detailed breakdown of its performance metrics and pricing structure. The announcement has generated significant engagement on Hacker News, indicating strong community interest in its capabilities.

Kimi K3 is positioned as an open frontier model for competitive benchmarking.

Agentic AI NEW GitHub Trending (daily)

GitHub launches multi-platform SDK to embed Copilot agent runtime in apps

GitHub released an official SDK for Python, TypeScript, Go, .NET, Java, and Rust to integrate Copilot's agentic workflows directly into custom applications. The toolkit exposes the same production-tested agent runtime used by Copilot CLI, allowing developers to programmatically invoke planning, tool invocation, and file edits without building orchestration from scratch. This enables embedding Copilot's capabilities into external services and internal tools using standard SDK patterns.

Official SDKs now available for Python, TS/JS, Go, .NET, Java, and Rust

AI / ML NEW The Register

Researcher poisons open-weight AI model for under $100

A researcher demonstrated that open-weight AI models can be compromised for less than $100 by exploiting their lack of verification mechanisms. The attack highlights a critical vulnerability where models are deployed based on trust rather than verifiable integrity. This low-cost vector suggests that current open-weight distributions may not be secure against targeted poisoning attempts.

Open-weight models lack inherent verification, making them vulnerable to trust-based attacks.

Trending on GitHub NEW GitHub Trending (daily)

PrismML Bonsai Demo: Local 1-bit/27B VLMs with Agentic Tools

The Bonsai-demo repo enables running Bonsai language models locally across Mac Metal, Linux/Windows CUDA/Vulkan/ROCm, and CPU targets. The updated Bonsai 27B family introduces vision-language capabilities, native OpenAI-style tool calling, and MCP server support. It also features a reasoning model with configurable effort and context windows exceeding 256k tokens.

Run Bonsai 1-bit and 27B models locally on diverse hardware including Mac Metal and GPU/CPU stacks.