Hi there, this is your daily ☕️ Devshot.
In today's Devshot:
📱 Google to require foldable-ready apps
✨ Gemini 4 Neon matches top coding models for 25% less
🤖 Netlify rebuilds its platform around AI agents
🚀 Pulumi launches AI agents that ship code
🔧 LinkedIn cuts Spark shuffle failures
⚡ Oxlint adds Rust type checker
Plus: 🎁 7 more stories you might like, 🛠️ 6 tutorials, 🧰 6 tools, and 📚 5 papers.
Google to require foldable-ready apps LINK
- Google will require Android apps to meet a baseline foldable-readiness standard on Google Play starting in April 2027, blocking non-compliant apps from publishing or updating and potentially lowering their store visibility.
- The Tier 3 bar demands apps fill the display without letterboxing or compatibility modes, support multi-window, and preserve scroll position, typed text, and media playback across rotation, folding, and window resizing, with higher tiers adding multi-pane layouts and peripheral support.
- New apps and updates must include three foldable screenshots from April 2027, while games get until August 2027 to hit Tier 3; that same month, apps must target Android 17, removing orientation-lock and resize-block opt-outs.
Gemini 4 Neon matches top coding models for 25% less LINK
- Google launched Gemini 4 Neon today on the paid Gemini API, AI Studio, Android Studio, and Antigravity, a mid-range model that matches Anthropic's flagship Opus 5.5 on coding benchmarks while undercutting its price.
- Neon scored 46.2% on Cognition's FrontierCode 2.0 against Opus 5.5's 47.3%, and runs about 1.9x faster than Gemini 4 Carbon at roughly 110 tokens/second, though first-token latency reaches about 57 seconds at the xhigh setting.
- Pricing runs $3 per million input and $15 per million output tokens, 25% below Opus 5.5 but 50% above Sol and Sonnet, so Google says Neon must use roughly a third fewer tokens than either rival to erase that premium.
Netlify rebuilds its platform around AI agents LINK
- Netlify has rebuilt its platform around AI agents, letting users describe an app and have Agent Runners build it fullstack, with Connectors to pull in external data and Netlify Server for long-running backends.
- Connectors let agents reach your Linear, Notion, GitHub, Stripe, Salesforce or Snowflake accounts via a URL and per-user authorization, so the agent reads context and takes actions rather than you pasting it into prompts; access stays personal and reusable across projects.
- Netlify Server gives apps a long-lived Node.js process running Express, Hono or Fastify that handles concurrent requests and WebSockets without blocking, is free during early access, and existing customers must join a waitlist while the new experience rolls out.
Pulumi launches AI agents that ship code LINK
- Pulumi has launched two products in public preview, Pulumi Delivery and Pulumi Agents, that let AI agents handle more of releasing software and fixing it when something breaks, extending its infrastructure-as-code approach past the merge.
- Delivery lets engineers and agents write release steps as code, and can gate shipping with Jev, a model from TypeSafe AI that estimates specific risks and halts the release if one exceeds a team-set threshold; Agents investigate outages, cut costs, and patch security with every change surfaced as reviewable code.
- Agents now handle 40% of operations on the Pulumi platform that people used to do, up from 15% in April, and the products could replace work currently done by Argo CD, GitHub Actions, and homegrown scripts, though Jev decision gates and agent setup need paid Pro or Enterprise plans.
LinkedIn cuts Spark shuffle failures LINK
- LinkedIn rolled out Apache Celeborn as a shared remote shuffle platform across most of its Spark and Flink workloads, separating intermediate data storage from compute and cutting daily shuffle fetch failures from roughly 750 to 20.
- Map tasks push intermediate data to dedicated Celeborn workers that store partitions independently of executors, backed by a five-node Raft control plane across availability zones, which drove a 20-fold reduction in task failures during infrastructure disruptions versus Magnet.
- Moving shuffle storage to NVMe SSDs cut shuffle duration by 40% and switching to Zstandard compression shrank the worker fleet by about a third, while overall per-task shuffle cost fell nearly 90%.
Oxlint adds Rust type checker LINK
- Oxlint now supports tsrs, the Rust port of TypeScript's type checker from Cloudflare and VoidZero, as a drop-in alternative to tsgolint for type-aware linting, with Cloudflare, Owner.com and OpenAI already running it.
- Running typechecking and type-aware linting through a single TypeScript analysis, tsrs reaches up to 7x faster speeds with 75% less memory than tsgolint, and is far quicker still versus running typescript-eslint separately.
- tsrs implements all 61 rules from tsgolint v7.0.2007 and preserves existing config, diagnostics, fixes, and suggestions, so you upgrade Oxlint and it switches automatically; the team asks you to report any differences in output, runtime, or memory.
Tutorials & Advice
> Data Contracts: publish machine-readable, verifiable descriptions of a dataset's shape and quality so people and AI agents can trust and use it correctly.
> S3 vs. ScyllaDB for tiny telemetry: What Red Canary learned the hard way: Red Canary cut cost and latency by storing huge volumes of tiny telemetry in the write-optimized database ScyllaDB instead of S3 storage and Postgres.
> How GraphRAG Works - A Deep Dive: GraphRAG links documents into a knowledge graph of connected entities, so an AI can trace relationships across many sources instead of retrieving isolated text chunks.
> Writable Derived State: Stop Synchronizing State: Instead of copying server values into client state and losing user edits on refresh, read from the server only when the client value is missing.
> How Can You Be Responsible for Code You Don't Understand, Especially When You Can No Longer Program?: Managing AI-written code means tracking business outcomes instead of code quality, because teams that stop coding lose the ability to understand or fix it.
> Traffic mix matters: Match load test traffic to real production patterns, since skewed key access can swing throughput by orders of magnitude in either direction.
Other news you might like
- We stopped a Railway migration 3 weeks out. How to know to stop or go? LINK
- NetWasm 0.10 Speeds Up C# on WebAssembly, Now Faster Than Mono AOT LINK
- AMD ROCm 10.2 Released With Latest Optimizations, Expanded Ryzen Laptop Support LINK
- Three 'e-waste' Nvidia V100s hit 100 tokens per second on Qwen 3.6 35B in a $1,042 homelab rig LINK
- Snapchat now lets developers build AR Lenses with AI assistants like Claude Code and Cursor LINK
- Solid 2.0 - The Long Hold LINK
- jQuery 4.1.0 and 3.8.0 Released: Passing the Torch LINK
Trending tools
beacon.sh: open-source persistent memory for AI coding agents, retaining context across sessions so they learn from past work instead of restarting each time LINK
Scalekit: connects AI agents to 30,000+ tools across 600+ apps via one MCP gateway, handling auth, scopes, and per-user permissions in production. LINK
Inferock: a local proxy for OpenAI, Anthropic, Gemini, and OpenRouter that tracks token usage, failures, and retries to catch billing overcharges. LINK
Plurai Simulation: tests AI agents in a simulated environment, surfacing failures and feeding insights back so performance improves with each run LINK
Beam: streams individual windows from a remote Mac as native local apps, each with its own Dock icon and ⌘Tab slot over an encrypted connection. LINK
S.Space: one workspace unifying your team's AI agents like Claude Code and Codex, with shared memory and handoffs so you stop re-explaining and copy-pasting. LINK
Trending papers & reports
AI honesty prompts barely make report-writing agents flag missing data, and even a structured "what's unresolved" section surfaced the gap only ~30% of the time, meaning simple instructions won't reliably stop agents from hiding flaws. LINK
Systems engineering foundations get a formal definition that treats a system's structure and behavior as one unified thing, giving engineers a consistent way to model, verify, and analyze complex systems. LINK
Social debt fixes ranked by IT professionals show that structural changes to how teams organize information and decisions beat attitude-based fixes, with one clear winner, consolidating scattered chats into a single channel. LINK
Fortran migration testing offers a standardized benchmark of 50 real codebases with built-in behavioral checks, letting companies fairly compare tools that convert aging scientific software to modern languages without breaking it. LINK
Coding agent guardrails catch the outdated software libraries AI coding tools pull from memory, blocking stale, vulnerable versions and steering agents to current releases in over 93% of cases, closing a real security gap. LINK
See you tomorrow for a new dose of ☕️ Devshot!
