---
title: "Latest reads"
description: "The engineering internet, summarised so you can actually read it."
---

# Latest reads
> The engineering internet, summarised so you can actually read it.

## Articles

### [NVIDIA Nemotron 3.5 Super VL Powers Specialized Agents with Multimodal Reasoning](https://yomu.fyi/post/nvidia-nemotron-3-5-super-vl-powers-specialized-agents-with-multimodal.md)
- Company: [NVIDIA](https://yomu.fyi/company/nvidia-developer-blog.md)
- Author: Michelle Horton
- Published: Oct 15, 2026
- Type: Announcement

NVIDIA released Nemotron 3.5 Super VL, an open 120B-parameter multimodal model for specialized enterprise agents. Its mixture-of-experts architecture activates 12B parameters per token, balancing capacity and inference efficiency. The model combines reasoning, instruction following, tool use, and understanding of text, images, documents, and video for tasks including coding, web research, document intelligence, and cybersecurity. On a single GPU, it achieves up to 5x higher throughput for agentic tasks than competing open models, and it scores higher than the larger Nemotron 3 Ultra on Hermes Bench and LangChain evaluations. Early adopters report cost reductions of 40x for credit processing and 10x lower latency for receipt parsing, and organizations can post-train the model with NVIDIA NeMo for domain-specific workflows and deploy it with NVIDIA NIM and Dynamo, which serves up to 8x more users with 85% lower first-token latency.


### [OLTP is expensive, you can ClickHouse it.](https://yomu.fyi/post/oltp-is-expensive-you-can-clickhouse-it.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Published: Oct 15, 2026
- Type: Problem & solution

PostgreSQL’s ACID guarantees cause substantial write amplification: each write flushes WAL for durability, copies rows via MVCC for isolation, and updates every affected index, amplifying 100 bytes of row change into tens of kilobytes of I/O. Costs scale linearly with data size, and even sharded Postgres offers only linear scaling with added coordination overhead, so growing datasets drive rising bills. The post proposes offloading settled, rarely-changing data to ClickHouse, whose columnar compression shrinks datasets 10-30x and whose cloud offering costs about 5x less than comparable Postgres infrastructure while also reducing Postgres compute load through fewer indexes. ClickHouse builds indexes in background batches and runs aggregations at roughly 100x lower cost than Postgres. ClickHouse Cloud provides WalShadow for millisecond-latency CDC and pg\_clickhouse FDW, enabling an application pattern that cuts the cost of serving settled data by an order of magnitude.


### [Deploy with Void](https://yomu.fyi/post/deploy-with-void.md)
- Company: [Cloudflare](https://yomu.fyi/company/cloudflare.md)
- Author: Michael Dong
- Published: Oct 15, 2026
- Type: Announcement

VoidZero has open-sourced Void, a toolkit for building and deploying full-stack Vite applications to Cloudflare. The release includes three products: Void Platform, Void SDK, and Void CLI, which integrate into a single Vite workflow. The SDK provides file-based routing, database integration via imports, and end-to-end type safety from the database to the user interface. The CLI handles building, migrating, provisioning resources, and deploying with a single command, automatically inferring bindings like databases or AI services. The Platform offers a shared deployment environment with access controls and resource limits, enabling teams to prototype and ship internal apps in minutes.


### [IntelliJ IDEA’s Java and Kotlin for Cursor and VS Code Hits 1.0](https://yomu.fyi/post/intellij-idea-s-java-and-kotlin-for-cursor-and-vs-code-hits-1-0.md)
- Company: [Jetbrains](https://yomu.fyi/company/jetbrains.md)
- Author: Marco Behler
- Published: Oct 15, 2026
- Type: Announcement

IntelliJ IDEA’s Java and Kotlin language support extension has reached its 1.0 stable release, now available for VS Code and Cursor editors. The extension is powered by IntelliJ IDEA’s language server and debug adapter, providing smart code completion, navigation, code analysis, refactoring, and build tool support for Maven, Gradle, and Bazel. It also offers powerful debugging capabilities including emulated breakpoints and async stack traces, along with JUnit test runner support. The 1.0 release resulted from close collaboration with the Cursor team, making Java and Kotlin a first-class experience in Cursor’s Agents Window with dedicated support for nullability and Spring Framework specifics. Starting with 1.0, the extension requires an IntelliJ IDEA Ultimate subscription but provides a 30-day free trial, and future updates will enable agents to call the LSP directly for faster AI-assisted development, with announcements pending for purely agentic licensing.


### [Red Hat OpenShift Virtualization memory management overview](https://yomu.fyi/post/red-hat-openshift-virtualization-memory-management-overview.md)
- Company: [Red Hat](https://yomu.fyi/company/red-hat.md)
- Author: Fabian Deutsch
- Published: Oct 15, 2026
- Type: Explainer

The post overviews recent memory management evolutions in Red Hat OpenShift Virtualization, spanning from the Linux kernel to the orchestration planes. Historically, the platform uses a 1:1 memory request mapping, but data shows most VMs use only about 20% of allocated RAM, creating a delta that scales infrastructure with workload definitions rather than demand. Node-level optimizations include free page reporting (FPR), which asynchronously recovers pages freed by the guest OS without forcing allocation like ballooning, and KSM zero pages for guest-agnostic deduplication of zero-filled memory. At the cluster level, Multi-Gen LRU integrated with control groups enables swapping to fast secondary NVMe storage, and a PSI-aware descheduler proactively migrates VMs to relieve memory pressure across nodes. OpenShift Virtualization 4.22 provides an autopilot feature that automates configuration for generally available memory optimizations, while developer preview capabilities require manual opt-in activation by default.


### [Cloudflare and OpenAI: making AI discovery better for everyone](https://yomu.fyi/post/cloudflare-and-openai-making-ai-discovery-better-for-everyone.md)
- Company: [Cloudflare](https://yomu.fyi/company/cloudflare.md)
- Author: James Balfour (OpenAI)
- Published: Oct 15, 2026
- Type: Announcement

Cloudflare and OpenAI launched a research pilot in July 2026 to help ChatGPT users discover content more efficiently using network signals about page existence changes and potential harm. The pilot ran on a cohort of about one percent of Cloudflare’s network and has enabled OpenAI to index hundreds of millions of new web pages that are useful but sparsely linked. Cloudflare provides these discovery signals to OpenAI’s crawler while OpenAI independently evaluates and ranks pages by relevance and helpfulness for ChatGPT Search. The collaboration also tests signals that flag harmful pages like fake login forms so crawlers avoid them without penalizing the legitimate domain. The next phase will refine page selection test faster change signals and measure business outcomes for site owners whose content is discovered.


### [October 2026 newsletter](https://yomu.fyi/post/october-2026-newsletter.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Published: Oct 15, 2026
- Type: Announcement

The October 2026 ClickHouse newsletter announces CostBench results where ClickHouse Cloud delivered 752× better performance per dollar than Databricks and 438× to 512× better than BigQuery on real-time analytics workloads. The 26.9 release extends LIMIT with AFTER and UNTIL boundary conditions, adds incremental refresh for materialized views, and introduces CREATE TOKEN for time-limited credentials. Executable user-defined functions are now generally available on ClickHouse Cloud, supporting precompiled Rust, Go, C++, and JavaScript binaries with outbound network access and a deterministic flag. Featured community member Hoang Minh V has had over 100 pull requests merged since May, including a fast path for PromQL aggregations and fixes to TimeSeries reads. The newsletter also covers the official Apache Airflow provider for ClickHouse, the chDB durable layer for agent memory, and a Postgres round-up on memory safety and local NVMe storage.


### [How Netflix Made AI Coding Agents More Efficient with JetBrains IDE Intelligence](https://yomu.fyi/post/how-netflix-made-ai-coding-agents-more-efficient-with-jetbrains-ide-in.md)
- Company: [Jetbrains](https://yomu.fyi/company/jetbrains.md)
- Author: Anastasia Kazakova
- Published: Oct 15, 2026
- Type: Benchmark

Netflix integrated JetBrains’ headless code-intelligence engine into coding agents Claude Code and Codex to address agents’ struggles navigating Java/Kotlin code across compiled JAR boundaries. The engine provides deterministic indexing of project source files and libraries, and Netflix built a daemon and CLI to make the LSP callable by agents, keeping the index warm across repeated calls. In benchmarks on two real codebases—an internal Spring Boot service and the open-source DGS Framework—agents using JetBrains intelligence were typically 20-30% more efficient in tokens and time, with up to 65% improvement for stronger models like Opus 5 at high effort. Correctness held steady or improved across all configurations, and the gains were consistent across every model and agent tested. Netflix plans to extend the same benchmark methodology to Python, TypeScript, and Go, as coding agents for those languages face similar problems.


### [MPS 2026.2 Release Candidate Available](https://yomu.fyi/post/mps-2026-2-release-candidate-available.md)
- Company: [Jetbrains](https://yomu.fyi/company/jetbrains.md)
- Author: Vaclav Pech
- Published: Oct 15, 2026
- Type: Announcement

JetBrains has released the MPS 2026.2 Release Candidate, a language workbench for creating domain-specific languages, with several key improvements over the previous EAP1 build. The RC adds Linux ARM64 support with a bundled JetBrains Java Runtime and the ability to build standalone MPS-based applications for that architecture. Build Language now supports importing files from TAR archives, preserving symbolic links and file permissions, which is useful for packaging platform-specific plugins and native libraries. The Projectional Agent Toolkit receives a major performance boost with token usage reductions exceeding 50% in tests using Claude Code, plus new MCP tools for typesystem debugging, intentions, and quick-fixes. Archive packaging preserves symbolic links and uses .tar.gz for macOS to avoid duplicate native libraries, while the Import dialog now respects user selections and the Diff dialog has visual improvements.


### [What Makes a Great KotlinConf Proposal?](https://yomu.fyi/post/what-makes-a-great-kotlinconf-proposal.md)
- Company: [Jetbrains](https://yomu.fyi/company/jetbrains.md)
- Author: Jelena Ilic
- Published: Oct 15, 2026
- Type: Explainer

The KotlinConf 2027 Call for Papers is open, and the post offers guidance for crafting a strong proposal, emphasizing that a valuable talk can come from a simple lesson learned rather than a groundbreaking discovery. Reviewers first notice the overall care, including a clear title and focused abstract, and they can detect AI-generated content, so authenticity and a personal voice are crucial. A good proposal starts with a central story that identifies a problem, what was tried, and what was learned, even from a small project, and it should be tested with colleagues. The title should communicate the topic and benefit, avoiding vagueness, while the abstract should provide two to three paragraphs of meaningful detail that outline the talk’s journey and include concrete takeaways. The post advises choosing the right format between lightning and regular sessions, and reviewing the proposal for clarity and authenticity before submitting.


### [Apertus 1.5 Release Notes](https://yomu.fyi/post/apertus-1-5-release-notes.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Hanna Yukhymenko, Oleg Lavrovsky, Y Metz
- Published: Oct 15, 2026
- Type: Announcement

Apertus 1.5 extends the 8B and 70B open-weight models with image and audio understanding, a thinking mode, and a 262,144-token context window, released under Apache 2.0. The models are a continued pretraining of Apertus 1.0, adding 4 trillion tokens to the 8B model and 2 trillion to the 70B model, for totals of 19T and 17T respectively. Multimodality uses an early-fusion tokenizer with Emu3.5 for images and WavTokenizer for audio, keeping the architecture unchanged and training at text-only speed. Post-training includes SFT, long-context SFT, reinforcement learning with verifiable rewards, DPO, and charter alignment via SDPO, with all data and recipes publicly documented. Evaluation results show competitive performance on reasoning, coding, and multilingual tasks, with the 8B model matching GPT-OSS-20B on translation and outperforming Claude Haiku 4.5, reported transparently via the EvalEval Coalition’s EEE schema.


### [How a new branch gets its config, migrations, and functions](https://yomu.fyi/post/how-a-new-branch-gets-its-config-migrations-and-functions.md)
- Company: [Supabase](https://yomu.fyi/company/supabase.md)
- Author: qiao
- Published: Oct 15, 2026
- Type: Explainer

Supabase branches are independent projects that require settings, schema, and edge functions to be set up before use. When a branch is created, configuration copies main project settings including Postgres, Auth, and Realtime config, then applies them with a single Postgres restart via walg\_enable; Auth secrets do not carry over. Migrations come from the git repo for GitHub-connected branches or from the main project’s migration history for gitless branches; each file runs as one transaction in pipeline mode. Edge functions are copied from the main project at creation in parallel with configuration, and on each push only changed functions are redeployed using Brotli compression and SHA-256 checksums. If a step fails, it shows as DEAD and stops dependent steps, while PAUSED indicates a warning that does not block subsequent steps.


### [Sync real-time Stripe data to Supabase with Stripe Data Pipeline](https://yomu.fyi/post/sync-real-time-stripe-data-to-supabase-with-stripe-data-pipeline.md)
- Company: [Supabase](https://yomu.fyi/company/supabase.md)
- Author: gregor\_vand, saba\_pochkhua, staaldraad
- Published: Oct 15, 2026
- Type: Announcement

Stripe’s Data Pipeline now adds Postgres as a destination, enabling developers to sync Stripe data directly into their Supabase project without maintaining webhooks or working around API rate limits. Setup occurs entirely in the Stripe Dashboard: under Reporting, Data management, users select Data Pipeline, choose Postgres and Supabase, and authorize the connection. Stripe backfills historical records and continuously syncs charges, refunds, disputes, customers, subscriptions, and invoices into a dedicated stripe\_sdp schema in the user’s Postgres database. A dedicated stripe\_sdp role with scoped permissions, JWT direct connections using short-lived tokens, and enforced SSL ensure the database password is never exposed to Stripe. Developers can then query the synced tables with standard SQL, join them with application data, and implement features like MRR calculations, churn analysis, or subscription-based feature gating without extra API calls.


### [Reducing Qwen training time by 6x on AI Runtime](https://yomu.fyi/post/reducing-qwen-training-time-by-6x-on-ai-runtime.md)
- Company: [Databricks](https://yomu.fyi/company/databricks.md)
- Author: Bruce Fontaine, Pardis Beikzadeh
- Published: Oct 15, 2026
- Type: Tutorial

The post describes systematic performance optimization to reduce fine-tuning time for the Qwen 3.8 27B model on AI Runtime from 7 hours 44 minutes to 1 hour 17 minutes, a 6x improvement. The baseline showed 98% GPU utilization, but torch traces revealed the CPU was stalling the GPU with too many small operations. Installing manually fused kernels such as flash-linear-attention and causal-conv1d reduced the number of GPU operations by 15x and improved training speed by 3.7x. Further gains came from switching from 2D padded batches to 1D packed tensors with correct position\_ids, which reduced the number of batches by 36% and eliminated dynamic shape recompilation. The post emphasizes measuring what matters beyond GPU utilization and leveraging pre-built community kernels and efficient batching.


### [Unlock powerful insights and forecasts with Predictive AI Functions](https://yomu.fyi/post/unlock-powerful-insights-and-forecasts-with-predictive-ai-functions.md)
- Company: [Databricks](https://yomu.fyi/company/databricks.md)
- Author: Jonathan Brito, Gary Shambat
- Published: Oct 15, 2026
- Type: Announcement

Data teams frequently field questions about future sales, metric movements, and anomalous data, but answering them typically requires data science teams to build and manage models, creating a bottleneck. Predictive AI Functions, now in Beta, remove that bottleneck by providing SQL-native functions for forecasting, driver analysis, and anomaly detection, backed by state-of-the-art ML models maintained centrally by Databricks. The launch includes ai\_forecast for time series projection, ai\_top\_drivers for segment driver analysis, ai\_detect\_anomalies for outlier detection, and ai\_predict\_value and ai\_predict\_class for regression and classification. These functions can be called directly in SQL, invoked through Genie with plain-language questions, or embedded in dashboards, delivering consistent, high-quality answers without infrastructure management. Benchmarking on fev-bench shows ai\_forecast outperforms forecasting functions from leading cloud data warehouses, achieving industry-leading accuracy out of the box with no tuning required.


### [Twilio Video Media SDK Powers Smarter Customer Engagement](https://yomu.fyi/post/twilio-video-media-sdk-powers-smarter-customer-engagement.md)
- Company: [Twilio](https://yomu.fyi/company/twilio.md)
- Author: Andy O'Dower
- Published: Oct 15, 2026
- Type: Announcement

Twilio launched the Twilio Video Media SDK in Public Beta, a server-side software package that connects applications to a Twilio Video Room as a secure virtual participant, streaming real-time audio and video to backend systems. This launch supports AI-enabled customer engagement by allowing teams to use consistent AI logic across chat, voice, and video channels while keeping sensitive data secure. The SDK enables use cases such as multimodal AI assistants that see and hear users, ambient healthcare documentation, remote verification with computer vision, and server-side compliance checks. By routing individual media tracks to a secure backend, the SDK supports HIPAA and GDPR requirements without client-side rendering. Early adopter GLOBO Language Solutions uses the SDK to enhance real-time quality and AI integration for patient care in healthcare environments.


### [How to Build a Voice AI Agent with OpenAI's GPT-Live, Twilio Agent Connect, and TypeScript](https://yomu.fyi/post/how-to-build-a-voice-ai-agent-with-openai-s-gpt-live-twilio-agent-conn-typescript.md)
- Company: [Twilio](https://yomu.fyi/company/twilio.md)
- Author: Wen Zhu, Xinghao Huang
- Published: Oct 15, 2026
- Type: Tutorial

The post addresses the common problem of voice AI demos failing in production by introducing a solution using OpenAI’s GPT-Live and Twilio Agent Connect. GPT-Live is a full-duplex speech-to-speech model that separates conversation from background work, and Twilio Agent Connect provides a GPTLiveProvider to bridge Twilio Media Streams to GPT-Live without custom WebSocket plumbing. The tutorial walks through building a TypeScript application that connects inbound callers to a GPT-Live-powered voice agent, including setting up a Fastify server, configuring the session with a welcome instruction, and wiring tool calls via Responses delegation using a get\_weather example. It also covers placing outbound calls using the same provider and demonstrates features like per-call customization and transcript logging. The result is a production-ready voice agent that can greet callers, answer questions, and execute backend functions, with the same pattern applicable to scenarios like appointment booking or order status lookups.


### [Building AI with MongoDB and Patronus AI](https://yomu.fyi/post/building-ai-with-mongodb-and-patronus-ai.md)
- Company: [MongoDB](https://yomu.fyi/company/mongodb.md)
- Published: Oct 15, 2026
- Type: Explainer

Large language models can be unreliable, with serious consequences in regulated industries. Patronus AI, an industry-first automated evaluation platform, enables engineers to score and benchmark LLM performance on real-world scenarios, generate adversarial test cases, and monitor hallucinations and PII. Research using FinanceBench found that state-of-the-art LLMs hallucinated or incorrectly answered up to 81% of financial analysts’ questions, even with RAG and external context. Patronus partnered with MongoDB Atlas to offer a RAG evaluation API and published a 10-minute guide demonstrating evaluation of a RAG system built with LlamaIndex and Atlas Vector Search on an SEC 10-K filing. Developers can improve RAG performance by adjusting indexes, chunking sizes, prompts, or fine-tuning embedding models, and should continuously test to maintain improvements.


### [What is agentic identity? Securing AI agents in 2026](https://yomu.fyi/post/what-is-agentic-identity-securing-ai-agents-in-2026.md)
- Company: [Twilio](https://yomu.fyi/company/twilio.md)
- Author: Ravleen Kaur
- Published: Oct 15, 2026
- Type: Explainer

AI agents are shifting from passive copilots to autonomous actors that book tickets, update calendars, and send messages without step-by-step human intervention. This shift introduces operational risks because traditional security controls designed for deterministic software cannot handle non-deterministic agent behavior, leading to over-permissioned agents, rogue behavior, and lack of real-time monitoring. Agentic identity is proposed as a governance layer built on three pillars: authentication (cryptographic keys), authorization (contextual permissions), and context (delegated link to end user). The framework includes short-lived session tokens, user context binding via token exchange, hard operational bounds on tool calls, human-in-the-loop gateways for high-risk actions, verified communication endpoints, and centralized logging with kill switches. By implementing these controls, organizations can deploy autonomous agents safely while maintaining visibility, containment, and auditability across all agent actions.


### [Announcing DPoP Support for Google OAuth: Securing Refresh Tokens](https://yomu.fyi/post/announcing-dpop-support-for-google-oauth-securing-refresh-tokens.md)
- Company: [Google](https://yomu.fyi/company/google.md)
- Author: Sergei Akulich, Brian Daugherty
- Published: Oct 15, 2026
- Type: Announcement

Google announced DPoP (RFC 9449) support for its OAuth platform, specifically binding refresh tokens to a client-generated asymmetric keypair to neutralize token theft and replay attacks from Infostealer malware and AiTM threats. Access tokens remain standard bearer tokens to avoid the per-request asymmetric cryptographic verification overhead on high-QPS APIs like Gmail and Drive, delivering a pragmatic tradeoff between security and performance. The implementation follows RFC 9700 best practices for sender-constrained refresh tokens and is recommended for Backend-for-Frontend (BFF) and Token-Mediating Backend architectures. DPoP operates at the HTTP header level, making it resilient in multi-hop environments such as Model Context Protocol (MCP) agent chains, though multi-agent delegation requires Google Cloud Secure Token Service (STS) to preserve key isolation. Complementary protections against cookie theft are offered by Device Bound Session Credentials (DBSC), which cryptographically binds browser session cookies to device hardware for a defense-in-depth strategy.


[Newer posts](https://yomu.fyi/posts.md) · [Older posts](https://yomu.fyi/posts/page/3.md)