Every post, summarised

Sep 3, 2026

• Spotify

  Spotify’s Backstage Portal cut my Claude Code token usage by 90%

  Claude Code often spends frontier-model tokens on file I/O rather than reasoning, prompting a routing approach that sends predictable work to cheaper models. The post implements this with two public AiKA Modes in Spotify’s Backstage Portal: bulk-reader summarizes large file sets, and code-writer generates patterned files from a specification and reference file, using Gemini 2.5 Flash in the examples. A Claude Code plugin called shunt enforces delegation through PreToolUse hooks, scripts, and skills; reads above a configurable threshold, defaulting to 350 lines, are blocked and redirected. In tests on a Java monorepo, bulk-read saved around 90% of Claude’s tokens, while generated code can go directly to disk. The design does not cover editing or reasoning, and 10–30-second delegation latency makes small tasks counterproductive.

• Meta

  ZGateway: Learnings from Putting a Proxy in Front of ZippyDB

  ZGateway is a stateless proxy tier introduced to unify traffic from ZippyDB’s large, diverse client fleet and address the reliability limits of direct access. By terminating TLS, authorizing requests, applying admission control and shaping, and routing through regional tiers, it replaces a many-to-many connection mesh with bounded client-to-gateway and gateway-to-server hops. Its embedded thick C++ client batches and coalesces shard requests, while a shared pipeline supports pure-proxy and read-through-cache tiers; per-use-case metrics, traces, and quota usage provide operational visibility. The gateway handles more than 1 billion operations per second, carries about 40% of ZippyDB traffic, and adds about 6% computational overhead to an average use case, with traffic projected to exceed 60%; production work also consolidated transaction state handling across 100% of transaction traffic without a reliability regression.

• NVIDIA

  NVIDIA PAIR Virtual Inference Router Expands Available Compute on Your Local Network

  NVIDIA Personal AI Router (PAIR) addresses the GPU bottleneck created when multi-agent workflows and concurrent local sessions generate many independent inference requests. It acts as a virtual inference router rather than an inference engine, proxying compatible Ollama and LM Studio interfaces, discovering and securely pairing local systems, and scheduling each request onto one eligible node based on readiness, engine state, model availability, workload, and GPU utilization. The agent keeps one familiar connection, while PAIR streams responses back and exposes placement through Jobs and metrics views; it does not merge GPUs, pool VRAM, or split a single request across machines. In an unofficial Hermes five-subagent demonstration, a three-device cluster averaged 8 minutes 48 seconds versus 18 minutes on one RTX Spark laptop, with results explicitly dependent on configuration and workload.

• Stack Overflow

  Elevating security, control, and accessibility: Stack Internal 2026.6

  Stack Internal 2026.6 is a release focused on administrative security, API control, developer portal integration, knowledge-health automation, and accessibility for enterprise AI workflows. Its new Admin Security page supports X-API-Key headers for API v2.3 requests, inactivity windows, per-application daily limits up to 10,000 requests per day, and service-credential rotation with non-displayable secrets. The release adds scheduled expiration for Announcement articles, generally available community broadcasts, expanded custom award limits, tag preference and synonym APIs, richer user data, and a Backstage plugin update supporting the New Frontend System alongside the legacy architecture. Accessibility work improves screen-reader announcements, keyboard and reflow navigation, focus management, high-zoom layouts, contrast, and semantic markup, while additional fixes address tags, suggested edits, and Slack rendering.

• Hugging Face

  NeoMME: an efficient Multimodal-native and Multilingual Encoder

  NeoMME is a family of 260M and 800M multilingual multimodal encoders designed to represent text and images without a separate pretrained vision tower or causal language model. A single bidirectional Transformer processes text tokens and raw 32×32 image patches, while pretraining from scratch uses a masked discrete-diffusion text-denoising objective with high masking to encourage image-grounded reconstruction. The retrieval variant adds jointly trained dense mean-pooled and 128-dimensional late-interaction heads, returning both in one forward pass for page-image retrieval. On ViDoRe v3, the 260M model reaches 0.523 and both sizes lie on the model-size Pareto frontier; at 2048×2048, it encodes about 51 pages per second on an NVIDIA L40S, roughly twice ColModernVBERT's throughput. Hierarchical token pooling and asymmetric quantization reduce late-interaction storage from roughly 1.5 MB to 6 kB per page, a 255× reduction, while retaining more than 95% of baseline nDCG@10.

• Jetbrains

  Learning to Code in the Age of AI: Advice From a Top Udemy Instructor

  An interview with Python educator Ardit Sulce examines how beginners should learn programming as AI changes development and teaching. Sulce says AI now handles many small questions, while teachers provide curated learning paths, judgment, and context; he also argues that students should preserve productive struggle while outsourcing setup friction. He recommends evaluating understanding through code explanation, review, and debugging rather than treating working code as proof, and identifies reading unfamiliar code, debugging, and system understanding as important junior skills. His proposed three-stage path moves from writing syntax by hand in a professional IDE, to using AI as a tutor, and finally to using it as a co-worker while reviewing increasingly complex projects.

• Jetbrains

  Kotlin Toolchain 0.12: Multiplatform Library Publishing, Wasm Apps, and More

  Kotlin Toolchain 0.12.0 extends its multiplatform build and publishing capabilities with multiplatform library publication, preview support for Wasm applications, command-line Compose Hot Reload, and broader IDE and terminal improvements. Libraries use the existing publishing configuration and produce a common API, per-platform artifacts, sources, module metadata, plus commonized and platform-specific cinterop bindings; Compose Multiplatform resources are not yet included. wasm-js/app modules can be built into ready-to-use web applications, run with kotlin run, customized through resources, and supplied transitive npm dependencies from Kotlin Multiplatform libraries. The release also reduces default Maven Central files, supports Compose Hot Reload through an MCP server, and recommends project-root-relative // dependencies while retaining the older syntax. It raises minimum compilation requirements to JDK 17 and Kotlin 2.2.20, updates bundled versions including Kotlin 2.4.10 and JDK 25, and recommends IntelliJ IDEA 2026.2.1 or newer for the latest features.

• Stack Overflow

  The economics of agent scale: tokens, ROI, and building platforms for AI-first teams (Part 2)

  The conversation examines the economics and platform implications of operating dozens or hundreds of AI agents rather than a small number of human-driven workflows. Andi Gutmans argues that many current models are already sufficient for automation, shifting the optimization problem to the complete system: model, harness, context, personalization, and evaluation. The goal is the least sophisticated model and smallest viable context that produce a reliable outcome at the lowest cost, with better trajectories reducing reasoning loops and token processing; Google’s stated approach is to hill climb models and data together. The discussion also addresses cost governance, search and observability for agents as a distinct persona, human-in-the-loop handling, and the continuing importance of computer-science fundamentals, agentic fluency, and business judgment.

• Red Hat

  Evaluate AI agents with IBM CLEAR & EvalHub on OpenShift AI

  Reliable AI agents can fail in production through incorrect tool use, off-topic reasoning, or plausible but incorrect answers that unit tests, static benchmarks, and logs may not expose. This tutorial builds an eval-driven trace-evaluation pipeline on Red Hat OpenShift AI 3.4 or later, using MLflow to store structured agent traces and evaluation artifacts, IBM CLEAR to score reasoning quality, relevance, and tool usage with an LLM judge, and EvalHub to orchestrate OpenShift jobs. It uploads eight prerecorded multistep research-agent traces, each containing at least 30 spans, then verifies them in MLflow before deploying EvalHub, registering the ibm-clear provider, and configuring judge-model credentials. Standard mode evaluates steps holistically, while SPARC mode gives tool calls separate scoring rows; the resulting workflow produces critiques and ranked failure patterns with frequency, severity, and step or agent attribution.

• Red Hat

  Extend Layer 2 networks into Red Hat OpenShift Virtualization with BGP and EVPN

  The article describes how to extend Layer 2 networks from a legacy data-center fabric into Red Hat OpenShift Virtualization using BGP and EVPN. It uses a three-node OpenShift lab with secondary NICs on a provider network, static IPv4 addressing, a VTEP CR for 172.19.0.0/24, FRRouting peering, and a Layer 2 primary CUDN with VNI 100. The configuration preserves imported VMs’ IP and MAC identities while maintaining isolation for tenant networks that may have overlapping address spaces. In the example, an external FRR router learns the VM MAC through EVPN and reaches 22.100.0.3; after the VM moves from node 3 to node 1, the MAC remains unchanged and the advertised VTEP changes from 172.19.0.102 to 172.19.0.100. The stated result is phased migration with Layer 2 adjacency to legacy application tiers and end-to-end tenant segmentation.

• MongoDB

  MongoDB VFS for LangChain Deep Agents: A Searchable Filesystem for Agents

  Modern AI agents are expected to complete real work, not just answer questions, which requires reliable filesystem navigation and updates. The MongoDB virtual filesystem (VFS) for LangChain Deep Agents connects this filesystem experience to MongoDB Atlas and an underlying object store like S3. It provides agents with a searchable workspace that persists across sessions, deployments, and sub-agents. Developers can use MongoDB Atlas as the data platform, including MongoDB Search, Vector Search, and hybrid search, without changing agent code. This architecture allows agents to plan, inspect source material, create artifacts, and resume work over hours or days without loading the entire corpus into context.

• MongoDB

  MongoDB Enterprise Advanced for AI-Ready Operational Insights in Manufacturing

  Unplanned manufacturing downtime costs $2.3 million per hour in the automotive sector, and a workforce shortage means 26% of existing workers will retire by 2030. Existing predictive maintenance systems operate in isolation, increasing mean equipment recovery times from 49 to 81 minutes per incident. MongoDB Enterprise Advanced provides a unified data platform for multi-agent AI systems that combine sensor telemetry, vector embeddings, agent memory, and operational records in a single database. Its document model, hybrid search, time series collections, and change streams enable agents to perform semantic retrieval, keyword search, and real-time stream processing. A multi-agent workflow includes a monitoring agent that detects anomalies, a diagnostics agent that retrieves historical patterns, a planning agent that drafts work orders with human-in-the-loop review, and a scheduling agent that checks production calendars and inventory, all on a single on-premise deployment with enterprise security and local embedding generation.

• Resend

  Welcoming Kiersti Esparza, our new Deliverability Engineer

  Resend introduces Kiersti Esparza as a Deliverability Engineer on its Trust & Safety Squad, where she will help scale sending infrastructure as the company grows. She brings nearly 20 years in email deliverability, including building teams at Marketo, consulting at Twilio, and deliverability engineering at Klaviyo. Esparza describes the work as a combination of interpreting datasets, understanding email technology, and adapting to evolving authentication in an ongoing contest between bad actors and receivers. She says Resend attracted her with developer-oriented infrastructure, hands-on foundational work, and a culture centered on transparency and partnership, while advising operators to master overlooked infrastructure because its scarcity creates leverage.

• Twilio

  Build a Video Chat App with ASP.NET Core and Angular using Twilio Video

  This tutorial builds a small group video chat application with an ASP.NET Core Web API and Angular single-page app using Twilio Programmable Video. The API binds typed Twilio credentials, mints short-lived access tokens locally with the API Secret, reads active rooms through the Twilio REST API, and pushes room-list changes over SignalR instead of browser polling. The Angular client uses the twilio-video library to capture camera and microphone media, connect to rooms, render participants, and provide device selectors. Local development runs the API alongside ng serve, while a Release MSBuild target runs npm ci and npm run build so dotnet publish places the SPA in the API's wwwroot. The resulting local test demonstrates a two-participant call and a live active-room count that updates when someone joins or leaves.

• Twilio

  4 must-ask questions before you delegate to your AI agent

  Automation roadmaps that prioritize ticket volume and low complexity can improve containment while lowering customer satisfaction when customers did not want a task automated. The post proposes four screening questions: how reversible a mistake is, whether the task touches money, health, or identity, whether the customer needs a transaction or to be heard, and whether the AI can access the necessary context. It cites a Twilio survey of 7,652 consumers, including 80% comfortable with appointment or reservation scheduling and 60% unwilling to have AI handle a medical inquiry. Tasks that fail the test can be split between AI and people: AI gathers information, verifies identity, retrieves history, and drafts a recommendation before a human decides, with Twilio Conversation Orchestrator handing conversation history and profile to Twilio Flex. The recommendation is to start with tasks where errors are relatively cheap and the data is available.

• Auth0

  Five Reasons Why Your Auth0 Permissions Are Not in your Access Token

  Auth0 access tokens can lack a permissions claim despite apparently configured RBAC, and the post identifies five silent causes. The authorization request may omit the API audience, the API’s “Add Permissions in the Access Token” setting may be disabled, or a permission-bearing role may not be assigned to the user, including at the relevant organization level. Developers may also be decoding an ID token rather than the API access token, whose aud identifies the API. Client Credentials tokens represent applications rather than users and typically carry granted scopes instead of a permissions array. For custom Post-Login claims, the post advises using a namespaced claim rather than colliding with Auth0’s reserved permissions claim.

• Hugging Face

  Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO Steps

  This guide presents a low-cost recipe for improving structured-output compliance in Liquid AI’s LFM2.5-350M, a task where parseable, schema-adherent responses determine whether an LLM can feed downstream systems. It fine-tunes the model with Group Relative Policy Optimization (GRPO) through TRL, using about 500 samples, 100 steps, three weighted rewards for format, field count, and JSON Schema validation, and a LoRA adapter training roughly 6M parameters. Prompt augmentation covers fenced-code-block instructions and top-level-array tasks, while evaluation uses IFStruct with llama.cpp on an OpenAI-compatible local server. On 2,000 IFStruct samples, the BF16 base model passes 22.6%, rising to 29.7% after tuning. JSON improves from 18.0% to 31.9%, bare-list compliance from 16.6% to 29.7%, while YAML changes little.

• Fly.io

  Your Agent Speaks MCP. Give It a Computer.

  Sprites are presented as disposable cloud computers for running coding agents and as tools agents can create for testing, debugging, benchmarking, and multi-service work. The post describes MCP as the transport and authorization layer, while progressive disclosure comes from plugins and skills that reveal capabilities only when needed; CLI, REST, and machine-readable /.sprite/llm.txt remain available. The hosted MCP server returns file reads as MCP resources and attaches safety annotations, marking read-only and destructive operations and flagging exec and service_start as crossing the Sprite boundary. Agents connect through sprites.dev/mcp and authenticate to a specific Fly.io organization, with configurable scoping, a default five-Sprite cap, and an mcp- name prefix. The design emphasizes durable filesystems, real networks, and near-zero idle cost instead of stateless sandboxes, while the post argues these properties make disposable computers practical for agent workflows.

• Hugging Face

  Give Your Coding Agents a Memory You Own

  funes is presented as a durable memory layer for coding agents that turns existing session traces into searchable, provenance-backed working memory across machines and agent types. A single binary adds recall and get tools to Claude Code, Codex, pi, or Hermes, builds an incremental local index, and automatically indexes completed turns. Its deterministic pipeline normalizes traces, chunks them, embeds them with a pinned local model, stores them in Lance, and combines vector and BM25 search with rank fusion, cross-encoder reranking, recency weighting, and neighboring chunks. Memories can remain local or be bound to a private Hugging Face dataset owned by the user, with credential redaction, secret scanning, local caching, and Hub-provided access control and versioning. In the handoff-vs-recall benchmark, recall was the cheapest option on both tasks, costing 8x less than a written handoff on one and 4x less on the other, while compaction succeeded on only one task.

• Hugging Face

  Training a coding model to paint watercolours with TRL and OpenEnv

  The project reproduces an earlier idea: training a language model to write JavaScript that paints watercolours with p5.brush, using TRL and OpenEnv. Its open pipeline includes a hand-rated reference pool, an RL environment, training scripts, trained models, and Hugging Face Jobs, Spaces, Inference Providers, and Hub artifacts. The environment renders sketches in headless Chromium, restricts the model to ten library methods, rejects cheats through a gate, and combines gate, length, pairwise-judge, and HPSv3 rewards. Three reward mixes were trained: judge-led (0.60/0.30), hps-led (0.30/0.60), and hps-only (0.00/0.90); the first two ran to step 110, while validation stopped at step 60. The post reports that both pairwise-judge runs learned and identifies the hand-rated pool as the bottleneck.