---
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

### [Optimizing Agent Reliability and Tool Calling Efficiency for Chip Design Using NVIDIA NeMo Gym](https://yomu.fyi/post/optimizing-agent-reliability-and-tool-calling-efficiency-for-chip-desi.md)
- Company: [NVIDIA](https://yomu.fyi/company/nvidia-developer-blog.md)
- Author: Tanya Lenz
- Published: Oct 14, 2026
- Type: Problem & solution

AI agents in chip design often produce plausible but incorrect artifacts, making reliability a key barrier to production adoption. NVIDIA NeMo Gym, an open library for evaluating models and agents using environments, solves this by diagnosing failure modes in multistage workflows through verification of final outputs against reference artifacts. Applied to the Siemens Fuse EDA AI Agent for standard-cell characterization, NeMo Gym uses a verifier that checks coverage, validity, relative validation, integrity, and accuracy of generated Liberty files against known references. After refining the agent’s skills and MCP tool descriptions based on trajectory analysis of failures, the composite benchmark score improved from approximately 55% to 95%, while agent turns fell from 155 to 53 and output tokens dropped from ~81K to 29K. This method generalizes to other engineering domains by emphasizing evaluation of final artifacts and domain-specific verifiers.


### [Multi-region disaster recovery for Amazon API Gateway WebSocket APIs](https://yomu.fyi/post/multi-region-disaster-recovery-for-amazon-api-gateway-websocket-apis.md)
- Company: [AWS](https://yomu.fyi/company/aws.md)
- Author: Shubham Dhoka
- Published: Oct 14, 2026
- Type: Tutorial

WebSocket connections for real-time applications like live auctions and collaborative editors are vulnerable to regional outages because each connection is tied to a single AWS Region. A regional failure severs all active connections and loses session state. This post presents an active-passive multi-region disaster recovery pattern that uses Amazon API Gateway WebSocket APIs, Route 53 failover routing, DynamoDB Global Tables for cross-Region state replication, a canary health probe, and client-side reconnection logic. The canary Lambda function tests the endpoint every minute and drives a CloudWatch alarm that triggers a Route 53 health check to mark the primary Region unhealthy. Clients detect connection drops via application-level heartbeat messages and reconnect with exponential backoff and full jitter to avoid thundering herd. After reconnection, clients retrieve the latest application state from the secondary Region’s DynamoDB replica and resume their session automatically.


### [vLLM vs. SGLang: Choosing the right inference engine for your workload](https://yomu.fyi/post/vllm-vs-sglang-choosing-the-right-inference-engine-for-your-workload.md)
- Company: [Red Hat](https://yomu.fyi/company/red-hat.md)
- Author: Cedric Clyburn
- Published: Oct 14, 2026
- Type: Explainer

Open weight models now process nearly 80% of production tokens on Vercel’s AI Gateway, up from under 10% in December 2025, and OpenRouter shows a similar climb. This surge drives teams to consider self-hosting via inference engines like vLLM and SGLang, since swapping backends usually only requires a new base URL for an OpenAI-compatible API. vLLM’s PagedAttention, introduced by UC Berkeley in 2023, splits the KV cache into fixed-size blocks to eliminate wasted contiguous memory reservation and increase GPU throughput. SGLang’s RadixAttention, from LMSYS in January 2024, stores computed KV cache in a radix tree keyed by token prefix for reuse across related requests like multi-turn chat and agents. Independent benchmarks from Modal show comparable performance on random prompts without prefix sharing, indicating that workload type and hardware should guide engine choice.


### [Configure Red Hat OpenShift Dev Spaces for enterprise security](https://yomu.fyi/post/configure-red-hat-openshift-dev-spaces-for-enterprise-security.md)
- Company: [Red Hat](https://yomu.fyi/company/red-hat.md)
- Author: Ajay Kanse
- Published: Oct 14, 2026
- Type: Tutorial

Red Hat OpenShift Dev Spaces can be hardened for enterprise use by centralizing SCM authentication with OAuth, deploying a private OpenVSX extension registry, and enabling IntelliJ IDEA in restricted networks. The SCM integration uses a dedicated OAuth application and the scm-oauth-config Secret in the operator namespace, while disabling the default CA bundle mount reduces etcd overhead but requires an explicit corporate CA ConfigMap with keys ca.crt and githost to prevent TLS failures. A private OpenVSX registry with PostgreSQL-only architecture replaces external extensions, and automated governance rejects prereleases, timestamp builds, oversized patches, and previews through version-aware dependency resolution and idempotent publication tracking. For JetBrains support in air-gapped clusters, administrators mirror IDE tarballs to an internal repository and set editorsDownloadUrls; developers connect via JetBrains Gateway using either dashboard- or gateway-initiated workflows, both relying on an existing oc login session. This configuration gives platform teams oversight of authentication, extensions, and IDE tooling.


### [Inside Hyperdrive: How we solved a rare Hyperdrive MySQL caching bug](https://yomu.fyi/post/inside-hyperdrive-how-we-solved-a-rare-hyperdrive-mysql-caching-bug.md)
- Company: [Cloudflare](https://yomu.fyi/company/cloudflare.md)
- Author: Pranoy Dutta
- Published: Oct 14, 2026
- Type: Incident / postmortem

Hyperdrive is Cloudflare’s distributed database connection pooler for Workers; its MySQL support became GA after a beta that uncovered over 20 bugs, including a rare caching issue. The most challenging bug caused consistently empty query responses 0.001% of the time on cache-enabled instances, requiring deep protocol analysis. The root cause was that the Client and Endpoint nodes share a data center cache but use different MySQL capability flags, so a cache entry written by one node could be encoded with capabilities that the other node cannot correctly translate. When the reading node encountered such a mismatched entry, it either sent an early EOF or hung, resulting in empty rows for the Worker. The fix was to always write a canonical version of the query result bytes encoded with Hyperdrive’s own capabilities, regardless of which node stored it, which resolved the issue and raised reliability to the 99.99% target.


### [Boost Your Pipeline: Top 6 CI/CD Integrations for Qodana](https://yomu.fyi/post/boost-your-pipeline-top-6-ci-cd-integrations-for-qodana.md)
- Company: [Jetbrains](https://yomu.fyi/company/jetbrains.md)
- Author: Kerry Beetge
- Published: Oct 14, 2026
- Type: Tutorial

Qodana brings IntelliJ IDEA’s inspection engine into CI/CD pipelines, automating code quality checks on every commit to catch issues before they reach production. The tutorial covers six integrations—GitHub Actions with its v2025.2 marketplace action and pull request integration, GitLab’s native CI/CD component that removes Docker complexity, TeamCity’s GUI-based native build runner with no YAML required, Jenkins via Docker-based Jenkinsfile for complex workflows, Azure Pipelines’ QodanaScan@2025 task displaying results in the Extensions tab, and CircleCI with a container-optimized orb. Each integration supports quality gates, baseline comparisons, and automated feedback, enabling teams to enforce consistent coding standards without disrupting existing workflows. Teams can select the integration that matches their current platform without requiring new tools or retraining. This shift-left approach reduces rework by catching defects early and transforms the pipeline into a quality guardian.


### [How Cloud-Based Agent Sandboxing Works in Air](https://yomu.fyi/post/how-cloud-based-agent-sandboxing-works-in-air.md)
- Company: [Jetbrains](https://yomu.fyi/company/jetbrains.md)
- Author: Dennis Ushakov
- Published: Oct 14, 2026
- Type: Explainer

AI agents require sandboxing to restrict file, network, and system access, preventing unintended modifications and data leakage. The post categorizes five agent execution modes into local (workspace, git worktree) and remote (SSH, Docker, Cloud), with remote modes providing stronger isolation by default. In JetBrains Air, each cloud agent task runs in its own VM with the agent inside a Docker container, and all network connections are routed through a proxy that allows only trusted domains by default. Secrets are stored encrypted and exposed as environment variables, and the agent operates within the developer’s repository permissions. Cloud-based execution reduces local setup requirements and gives organizations control over network access policies, with cloud agents currently available to organizational users via Air in IDEs and Air Teams.


### [Sizing Lakebase Search: lessons from a real app and 4 million passages](https://yomu.fyi/post/sizing-lakebase-search-lessons-from-a-real-app-and-4-million-passages.md)
- Company: [Databricks](https://yomu.fyi/company/databricks.md)
- Author: Philipp Tiefenbacher
- Published: Oct 14, 2026
- Type: Benchmark

Slide Hub, an internal Databricks app for slide search, uses Lakebase Postgres with Lakebase Search extensions (lakebase\_vector and lakebase\_text) to run hybrid vector and BM25 search on the same database as application data. The app stores 23k slides and indexes up to 4M passages from the MIRACL English corpus. Benchmarking shows that below 8 Compute Units (CU), Lakebase Postgres is the bottleneck; above 8 CU, the app’s worker count limits throughput. Candidate pool size significantly affects cold-start latency: 200 candidates preserves 90.5% of top-10 results while touching a third of the pages of 1,000 candidates. A 14-minute load test cost 0.13, with Lakebase compute billing only while active. For production, a 0.5 CU minimum costs ~ 31-40/month and avoids cold-start delays.


### [KotlinConf Recap: Lessons from Verifone’s Architect on Kotlin in Mission-Critical Payments](https://yomu.fyi/post/kotlinconf-recap-lessons-from-verifone-s-architect-on-kotlin-in-missio.md)
- Company: [Jetbrains](https://yomu.fyi/company/jetbrains.md)
- Author: Alina Dolgikh
- Published: Oct 14, 2026
- Type: Explainer

Verifone architect Yuri Geronimus presented at KotlinConf 2026 on using Kotlin’s type system to enforce payment specifications at compile time. He described six incident patterns where a single misused word caused outages, compliance failures, or financial loss, including a terminal fleet outage from a nullable mandatory field. His method uses value classes for spec nouns, validate factories at boundaries, non-null types, masked sensitive data by default, and state-specific classes to make illegal states unrepresentable. Geronimus advocates incremental migration from legacy Java and C++ systems using four parallel tracks rather than a rewrite, noting that payment terminals must operate offline for up to 40 days by regulation. The approach shifts reliability from developer discipline to compiler enforcement, enabling engineers to speak the business language and propose regulatory changes.


### [Lauterbach and JetBrains Announce TRACE32 Integration in CLion IDE](https://yomu.fyi/post/lauterbach-and-jetbrains-announce-trace32-integration-in-clion-ide.md)
- Company: [Jetbrains](https://yomu.fyi/company/jetbrains.md)
- Author: Artemy Pestretsov
- Published: Oct 14, 2026
- Type: Announcement

Lauterbach and JetBrains have integrated TRACE32 debugging tools with CLion, enabling embedded developers to debug applications on target hardware directly from the IDE. The integration uses the Debug Adapter Protocol (DAP) to connect CLion to TRACE32 through the TRACE32 Debug Adapter, allowing developers to control execution, step through code, and inspect variables while TRACE32 handles target hardware communication. Two workflows are supported: launching a new TRACE32 PowerView session from CLion with scripts to configure the connection, or attaching to an existing TRACE32 instance. The integration requires CLion 2026.2.4 or later, TRACE32 PowerView 09.2026 (or build 195402), and TRACE32 Debug Adapter version 0.1.2 or later, with detailed configuration examples in the Lauterbach knowledge base. This brings code editing, navigation, and hardware debugging into one workflow, and a free webinar on October 27, 2026 will demonstrate the integration in action.


### [Why Redshift Serverless can’t match ClickHouse Cloud for real-time analytics](https://yomu.fyi/post/why-redshift-serverless-can-t-match-clickhouse-cloud-for-real-time-ana.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Published: Oct 14, 2026
- Type: Benchmark

CostBench streamed 113.2 billion stock-market quotes into both systems while four aggregate queries ran every ten minutes and two drill-down queries ran every hour. ClickHouse asynchronous inserts order raw data and build pre-aggregations during ingestion, while Redshift relies on separate background sorting and materialized view refreshes. ClickHouse Cloud led in preparation cost ($28.69 versus 522.03), normalized query cost ( 0.05 versus $24.97), and accumulated query runtime (58.12 seconds versus 1.69 hours). The combined score shows ClickHouse Cloud delivered 1,996× better end-to-end performance per dollar in this continuous-ingestion comparison. The difference originates in the preparation path: ClickHouse makes data query-ready during inserts, whereas Redshift’s streaming, sorting, and refresh steps advance independently.


### [Introducing Private Network Gateway](https://yomu.fyi/post/introducing-private-network-gateway.md)
- Company: [Databricks](https://yomu.fyi/company/databricks.md)
- Author: Alex Esibov, Manish Bansal, Stanley Wang, Justin Ly, Vijay Raja
- Published: Oct 14, 2026
- Type: Announcement

Databricks announced Private Network Gateway (PNG) in public preview on Azure and private preview on AWS, addressing the networking friction that prevented teams from moving workloads to serverless compute. PNG provides a single managed gateway from serverless compute into a customer’s virtual network, eliminating the need for per-resource private endpoints, load balancers, and proxy VMs. Customers configure the gateway via a Network Connectivity Configuration (NCC) and delegate one subnet, with Databricks managing the gateway infrastructure while customers retain control over routing and security. PNG supports use cases such as ingesting from private databases via Lakeflow Connect, connecting AI agents to private MCP servers, using private package repositories, and routing egress through customer firewalls. The gateway offers two traffic modes for precise control or broad egress inspection and works alongside existing private connectivity without disruption.


### [Sentry's Flutter SDK 10.0: Streaming Traces and More Accurate Crash-Free Rates](https://yomu.fyi/post/sentry-s-flutter-sdk-10-0-streaming-traces-and-more-accurate-crash-fre.md)
- Company: [Sentry](https://yomu.fyi/company/sentry.md)
- Author: Gino Buenaflor
- Published: Oct 14, 2026
- Type: Announcement

Sentry released Flutter SDK 10.0, making streaming traces the default so spans reach Sentry as they finish, using Sentry.startSpan for custom work. Unhandled Flutter errors no longer mark the session as crashed, improving crash-free rate accuracy by keeping the session running until a real crash occurs. App startup on Android and iOS now has its own trace with breakdowns such as root-widget attachment and first-frame rasterization, and can be extended with SentryFlutter.extendAppStart. The release simplifies logs and metrics setup by removing separate enable flags, expands Dio network error capture to failures without HTTP responses, and respects Flutter’s SensitiveContent widgets in Replay and screenshots. Upgrading requires Dart 3.12, Flutter 3.44, Android API 26, and iOS 15, with native SDK installation via Swift Package Manager on iOS and macOS.


### [Anthropic opens Singapore office and signs MOU with the Government of Singapore](https://yomu.fyi/post/anthropic-opens-singapore-office-and-signs-mou-with-the-government-of.md)
- Company: [Anthropic](https://yomu.fyi/company/anthropic.md)
- Author: Anthropic
- Published: Oct 14, 2026
- Type: Announcement

Anthropic opened a new office in Singapore, its fifth in Asia-Pacific, and signed an MOU with the Singapore Ministry of Digital Development and Information on AI safety, security, and model evaluation. The MOU, signed by co-founder Ben Mann and Permanent Secretary Kai Fong Chng, also covers research on AI adoption and responsible use across public and private sectors. Claude already powers services used by millions daily through organizations like GovTech, DBS, Grab, and Sea; Grab’s BriX assistant saves 40,000 hours quarterly. Singapore ranks second globally in Claude.ai usage relative to population, at nearly six times expected levels, and over 1,500 GovTech engineers use Claude Code. The office is hiring and plans technical exchanges, AI skills strengthening, and support for small and medium businesses.


### [A local Supabase project per agent, with or without Docker](https://yomu.fyi/post/a-local-supabase-project-per-agent-with-or-without-docker.md)
- Company: [Supabase](https://yomu.fyi/company/supabase.md)
- Author: jgoux, wenbo
- Published: Oct 14, 2026
- Type: Announcement

The Supabase CLI v2.119.0 introduces an experimental native runtime that runs all 13 services as plain processes without Docker, reducing startup time by over 80% and idle memory by 89%. It also supports multiple local projects on one machine via git worktrees, each with its own Postgres, Storage, Edge Functions, and MCP server on auto-assigned ports. This allows each coding agent to have an independent database to reset and test against without affecting others, making a project per worktree practical on a laptop. Remote branches now inherit configuration from their base project and are ready nearly five times faster, with function bundles copied in parallel. The native runtime runs on Linux and macOS, and Docker remains the default when present, but the CLI automatically selects native when Docker is absent.


### [How we made the Supabase Select 2026 badge](https://yomu.fyi/post/how-we-made-the-supabase-select-2026-badge.md)
- Company: [Supabase](https://yomu.fyi/company/supabase.md)
- Author: sam\_meech\_ward
- Published: Oct 14, 2026
- Type: Tutorial

Supabase gave every Select 2026 attendee a programmable badge based on the Pimoroni Tufty 2350, with an RP2350 microcontroller, color screen, rechargeable battery, Wi-Fi, Bluetooth, and five buttons. A firmware bug drained all batteries during shipping, requiring bulk charging with 30-port USB hubs in 14 batches over 56 hours and individual firmware updates. The venue Wi-Fi failed, so the schedule was synced via a hotspot, two badges at a time, one minute each, totaling seven hours. Attendees have since repurposed the badge into projects including a live flight tracker, an arcade game about database vacuuming, a food finder app, a music note collection game, and a tiny on-device language model. The post provides resources for hacking the badge: a first-app guide, a Claude Code project kit, a virtual badge, guides for C or Rust firmware, and a contest with Supabase credits and conference passes.


### [Inside Minerva: building scalable observability for eBay’s monetization platform](https://yomu.fyi/post/inside-minerva-building-scalable-observability-for-ebay-s-monetization.md)
- Company: [Ebay](https://yomu.fyi/company/ebay.md)
- Author: Gautam Singh, Ravi Agarwal, Akshay Kamath
- Published: Oct 14, 2026
- Type: Problem & solution

eBay’s Monetization Team lacked a dedicated observability platform, causing fee behavior issues to often be discovered through customer complaints or post-transaction analysis. Data warehouse data arrived with delays, and engineers and business teams had to manually query multiple sources. Minerva solves this with an asynchronous observability pipeline: the fee-processing service places a small record on a local queue, a background producer publishes it to an event stream, and a consumer enriches the data before storing it in ClickHouse. ClickHouse’s columnar storage and MergeTree engine allow fast aggregation queries on records with nearly 250 columns; a proof of concept on 1.1 billion rows returned queries in 0.11 to 0.72 seconds. The design keeps observability off the critical fee path, prioritizing fee-processing availability while providing configurable, context-aware monitoring.


### [How to Encrypt LangGraph Checkpoints With MongoDB CSFLE](https://yomu.fyi/post/how-to-encrypt-langgraph-checkpoints-with-mongodb-csfle.md)
- Company: [MongoDB](https://yomu.fyi/company/mongodb.md)
- Published: Oct 14, 2026
- Type: Tutorial

LangGraph checkpoints persist agent state, including user messages and tool results, as plaintext BSON in MongoDB by default, leaving sensitive data accessible to anyone who can read the database. MongoDB Client-Side Field Level Encryption (CSFLE) encrypts data on the client before transmission, ensuring the database engine never sees plaintext values and protecting against privileged database users or data breaches. This post demonstrates implementing a custom CipherProtocol for LangGraph’s EncryptedSerializer that uses explicit CSFLE with randomized encryption, configured with AWS KMS and a data encryption key stored in a key vault collection for envelope encryption. The encrypted payload replaces the default Base64 encoding, while only thread IDs, checkpoint IDs, and namespaces remain unencrypted to support runtime lookups without exposing sensitive content. This approach aligns with enterprise security and compliance requirements for customer-facing AI agents, preserving resumable sessions and human-in-the-loop workflows while preventing database administrators or attackers from reading the state.


### [Design Email for Humans and Agents](https://yomu.fyi/post/design-email-for-humans-and-agents.md)
- Company: [Resend](https://yomu.fyi/company/resend.md)
- Author: Kiersti Esparza
- Published: Oct 14, 2026
- Type: Problem & solution

Apple Mail and Gmail now generate AI summaries of emails, while agents read messages to automate tasks like logging receipts or tracking deadlines. Summarizers focus on the first 100 to 150 characters, so the main point and call to action must lead the email. Semantic HTML with h1 headings, a descriptive title, lang attributes, and role=‘presentation’ on tables creates a machine-readable outline, yet the Email Markup Consortium found 74% of emails lack an h1 and 96% lack lang. Hidden text styled to be invisible can hijack summaries, as Forcepoint demonstrated, and similar attacks have targeted Gemini in Google Workspace; such practices also trigger security flags. Testing in Gmail, Apple Mail, and Outlook, along with verifying the plain text version, confirms that both humans and agents interpret the email correctly.


### [The Design Space of Tri-Modal Masked Diffusion Models](https://yomu.fyi/post/the-design-space-of-tri-modal-masked-diffusion-models.md)
- Company: [Apple](https://yomu.fyi/company/apple.md)
- Published: Oct 14, 2026
- Type: Announcement

Discrete diffusion models have emerged as strong alternatives to autoregressive language models, but prior approaches typically initialize a unimodal model and fine-tune for bimodal generation. This work diverges from that paradigm by introducing the first tri-modal Masked Diffusion Model pretrained from scratch jointly on text, image-text, and audio-text data, and systematically analyzing multimodal scaling laws, modality mixing ratios, noise schedules, and batch-size effects. A key contribution is a stochastic differential equation reparameterization that decouples physical and logical batch sizes, removing the need for batch size tuning reported in recent work. The authors pretrain a 3B-parameter model on 6.4 trillion tokens and provide optimized inference sampling defaults, achieving strong results on text generation, text-to-image, and text-to-speech tasks. This work represents the largest open systematic study of multimodal discrete diffusion models conducted to date.


[Newer posts](https://yomu.fyi/posts/page/3.md) · [Older posts](https://yomu.fyi/posts/page/5.md)