DEV Community

April Fools Challenge is live!!1 What's the most useless thing you can build? Captcha that feels trigonometry?📐 A vertical scroll bar that moves horizontally?↕️ Show us how weird you can get and you could be one of five winners of a tea pot (really). dev.to/devteam/join-o…

Switching between a work and gaming monitor layout multiple times a day gets old fast. This dev built wsm, a Rust CLI that saves monitor configurations as YAML profiles and restores them with a single command — with atomic layout changes to avoid flickering. { author: @OsirisFrik } dev.to/osirisfrik/wee…

Built a tiny Go CLI that talks to an AVR over WebSocket, then wired it into Home Assistant as a command_line switch. The hard part wasn't sending commands — it was making state verification reliable enough for automations. { author: @moniquelive } dev.to/moniquelive/in…

Pre-training and post-training (or SFT and RL). This recap covers why Reinforcement Learning is the hardest step to implement, what makes TPU Pods uniquely suited for it, and how MaxText 2.0 integrates the full stack. { author: Shir Meir Lador + @GoogleAI } dev.to/googleai/agent…

Standard semantic search gets you "close" results — not necessarily the right ones. This dev adds IBM's Docling for structured PDF chunking and a re-ranking model that assigns a true relevance score after retrieval, filtering out results that don't actually answer the query. { author: Glen Yu + @GoogleDevExpert } dev.to/gde/implementi…

Cloud Run Jobs vs. Cloud Batch. The difference comes down to abstraction vs. control. Cloud Run Jobs caps GPU tasks at 1 hour per instance; Cloud Batch supports multi-day runs across up to 8 GPUs with MPI and RDMA for tightly coupled HPC workloads. { author: Maciej Strzelczyk + @GoogleAI } dev.to/googleai/cloud…

A senior JavaScript dev using a multiplayer Bomberman clone as a real reason to finally learn Rust. The approach: a "mentor-review" loop with AI where the dev writes the code, not the AI. { author: @levytomer85 } dev.to/tomerl1/im-reb…

@julianharris Let's see it!

@ThePracticalDev How about an agent whose system prompt is to delete everything it creates after creating it

April Fools Challenge is live!!1 What's the most useless thing you can build? Captcha that feels trigonometry?📐 A vertical scroll bar that moves horizontally?↕️ Show us how weird you can get and you could be one of five winners of a tea pot (really). dev.to/devteam/join-o…

Every token that arrives triggers a full layout reflow in the DOM. At 50 tokens/second, that's dropped frames and a jittery scrollbar. This dev built ZeroJitter — a React library that renders streaming text to canvas via a Web Worker, with zero reflows and full accessibility support. { author: JAI } dev.to/jvoltci/zeroji…

A backend team used AI to convert a Python class to JavaScript — and it failed. Not because the translation was wrong, but because the AI didn't know how to handle a 5GB zip file in the browser without crashing it. This dev did, using IndexedDB as an async workaround. { author: @OlumideSamuel_ } dev.to/olumidesamuel_…

A look at what went into building a fully responsive CRT TV in pure CSS — including 3D lighting on a spinning cube, volume knobs that distort scanlines, and glass reflections made entirely from repeated dashed borders. { author: @ivorjetski } dev.to/ivorjetski/cod…

15 minutes before every meeting, this agent wakes up, scans your emails and Drive files for context, and delivers a briefing to Google Chat with a link to a full Google Doc. No Python scripts. Built entirely in natural language using Google Workspace Studio. { author: @aryanirani123 + @GoogleDevExpert } dev.to/gde/google-wor…

Since HuggingFace dropped native JAX support, running those models on TPUs got complicated. TorchAX bridges PyTorch and JAX using a special tensor subclass — no model rewrites needed. This dev covers JIT compilation, text generation with StaticCache, and tensor parallelism. { author: Ahmed Elnaggar + @GoogleDevExpert } dev.to/gde/run-any-hu…

Not every step in an AI workflow needs an LLM. This dev shows how using a custom BaseAgent instead of an LlmAgent for deterministic tasks cuts latency, eliminates token costs, and makes test results fully predictable. { author: Connie Leung + @GoogleDevExpert } dev.to/gde/stop-wasti…

Can you build a performant, IO-heavy app without async Rust? This dev explores epoll directly, then rewrites the result using Mio — achieving single-threaded, non-blocking networking with time-based triggers and no busy waiting. { author: @Szymongib } dev.to/szymongib/asyn…

Google's Agent Development Kit used to visually define a multi-agent pipeline that generates comic book-style HTML. Tested locally, then deployed to AWS Fargate. This dev walks through the cross-cloud setup step by step. { author: xbill + @GoogleDevExpert } dev.to/gde/cross-clou…

AI will make every architectural decision for you if you let it. Rate limiting, auth strategy, pagination — it'll just pick something. This dev walks through using plan mode to treat the AI's first output like a real design review, not a mandate. { author: Ava Barron } dev.to/abarron/vibe-w…

It started with building websites… then one thing kept blocking me: payments. That obstacle didn’t stop me; it changed my path in tech From freelance dev to technical writing, here’s how 👇🏻 Submitted for the 2026 WeCoded Challenge on @ThePracticalDev 💜 dev.to/hadil/how-a-pa…

After a few sleepless nights and constantly questioning my life decisions, I finally finished my new project: NoteRunway, built for the Notion MCP challenge on @ThePracticalDev NoteRunway helps you clean up and refine your workspace like it’s brand new. dev.to/georgekobaidze…

my article on building phntm.sh got featured on @ThePracticalDev this week. pretty wild for a build in public post about zero knowledge file transfer and a first Go project. if you missed it: dev.to/aliirz/i-built…