daily.dev · Popular · 实时热榜

TanStack Charts is a new JavaScript charting library using a grammar-of-graphics approach (similar to ggplot2 in R), treating charts as compositions of marks, scales, and mappings rather than fixed chart types. It supports ~55 mark types including bars, lines, areas, sankey diagrams, tree maps, radar charts, world maps, and more. Key features include SVG and canvas output, animations, tooltips, interactions (brushing, zooming, keyboard selection), export to SVG/PNG, and a three-level styling sys

Agentic engineering is the shift from AI chatbots that just answer questions to AI agents that get work done for your engineering teams. The difference sounds small, but it changes everything about how you design these systems. A chatbot just needs to sound right. An agent needs to be right, because it’s the one filing the pull request, spinning up the infrastructure, or pushing to production. That’s why the hard part isn’t the model, it’s the context around it. An agent that understands your se

A developer argues that explicit code is always preferable to magic abstractions, drawing on frustrations with an opaque JavaScript library. The post covers four pillars: readable-by-design code with named intermediate variables, static types as explicit contracts, traceable imports/exports, and the cognitive cost of implicit conventions. Practical rules include preferring explicit exports, using static types, avoiding cross-language implicit transformations, favoring named variables over clever

Open Knowledge Compiler (open-knowledge-compiler, Apache-2.0) is a new open-source tool that continuously compiles a Git repository — its code, pull requests, and optionally Jira — into an OKF-conformant knowledge base. It serves both humans (as a browsable Markdown wiki) and AI agents (via a read-only MCP server with structured graph queries). The compiler uses a six-stage pipeline: Collect, Extract, Normalize, Diff, Persist, and Emit, with PostgreSQL as the durable source of truth. Determinist

LongCat 2.0 is a 1.6 trillion parameter open-weights LLM from Meituan-backed AI lab LongCat, trained entirely on Chinese AI hardware (ASIC-based AI6 superpods) without NVIDIA GPUs. The model matches near state-of-the-art performance and introduces three key architectural innovations: (1) LongCat Sparse Attention with streaming-aware indexing, cross-layer indexing with distillation, and hierarchical coarse-to-fine retrieval; (2) N-gram embeddings allocating 135B parameters to learn multi-token ph

shiki v0.9.0 is out — shiki tasks — the tasks view as a scriptable CLI command: pending tasks across every notebook, urgency-sorted, with --overdue/–today/–all/–notebook filter… https://sazardev.github.io/shiki/changelog/0.9.0.html #rustlang #TUI #opensource https://sazardev.github.io/shiki/changelog/0.9.0.html

A solo developer shares a five-month retrospective on building lerd, an open source local PHP development environment for Linux (and now macOS) that uses rootless Podman containers instead of Docker Desktop. The project grew from 0 to 780+ GitHub stars, attracted 16 external contributors, and users in 40+ countries. Key features added include a debug window intercepting dumps and SQL queries, an MCP server compatible with major AI coding tools, Git worktree isolation with per-branch databases, a

Tech giants like Google, Meta, and Amazon made headlines with massive layoffs while simultaneously hiring thousands in strategic areas, resulting in net headcount increases. The narrative that AI is replacing workers en masse is contradicted by data: Google added 16,000 net employees over three years and Meta added 8,000. The layoffs largely reflect post-COVID over-hiring corrections rebranded as AI transformation. Amazon's internal AI projects illustrate the hidden costs of agentic systems — on

pdf-inspector is a fast Rust library for PDF classification and text extraction, available as a crate and with bindings for Python, Node.js, and browser WebAssembly. It detects whether a PDF is text-based, scanned, image-based, or mixed in 10–50ms by sampling content streams, then extracts text with position awareness and converts it to clean Markdown — all without OCR or ML models. Key features include multi-column layout detection, dual-mode table detection (rectangle-based and heuristic), CID

Japan's Ministry of Foreign Affairs has contacted the US Embassy twice in June to formally request that the Trump administration stop using anime and gaming characters — including Pokémon, Naruto, and Mario — in political social media posts without rights holders' consent. The concern centers on reputational harm to these franchises. The Pokémon Company previously denied permitting a DHS mashup of Ash Ketchum footage with ICE raid clips, and the White House has also shared videos blending milita

Working with AI agents in 2026 has shifted software engineering toward rapid context-switching and skimming LLM outputs, leaving less time for slow, deliberate thinking. The author argues this risks atrophying deep reasoning skills and recommends two countermeasures: writing in your own words (not with LLMs) to force genuine articulation of ideas, and reading dense non-fiction books slowly. Combining both — reading a book and then writing about it — is presented as the most effective way to keep

A developer shares their return to Sublime Text after years of using VS Code and Zed, arguing that modern software is bloated and over-engineered. Sublime Text is still actively maintained with a stable build from May 2025 and dev builds as recent as July. The post advocates for focused, single-purpose tools over feature-heavy, constantly-updating editors.

Soft delete sounds simple: add deletedAt and hide the record. Then the real questions arrive: Can its unique value be reused? Should deleted relationships still appear? Can the record be restored safely? Does “deleted” satisfy privacy requirements? Will every query apply the correct filter? Soft delete is a data-lifecycle decision, not just a decorator. Without clear rules, your database slowly fills with ghosts that still affect production logic. When do you choose hard delete instead? #NestJS

DuckDuckGo partnered with sunglasses brand Knockaround to release "Normal F**king Sunglasses" — a satirical product mocking AI-powered smart glasses. The glasses have no camera, microphone, AI, battery, or electronics of any kind. They sold out in under a week. The collaboration is a commentary on growing surveillance concerns around AI glasses, particularly following Meta's use of Ray-Ban Meta smart glasses footage to train its AI systems and the June 2026 launch of Meta Glasses.

The way we build & deploy software is changing with AI agents & workflows, the Agentic Engineering way. Developers who understand how to use AI have an added advantage today. Well, already many of you might already be using many AI tools, AI agents and workflows to automate your SDLC. That’s a good move. What I have noticed is, many developers use AI tools to code but you need to go beyond that. See how you can automate the whole SDLC and cut off the manual hands-offs between different SDLC phas

A long-time Windows user reflects on why they haven't switched to Linux despite acknowledging its significant improvements. Key reasons include Windows' superior app compatibility, more familiar user experience, and the hassle of choosing a Linux distribution. The author appreciates Linux's privacy benefits and customizability but finds the switching cost too high when Windows still meets their needs. They leave open the possibility of switching in the future if Windows becomes frustrating enoug

As AI agents generate code at unprecedented speed, the real bottleneck has shifted from writing code to understanding and taking accountability for it. Reviewing AI-generated code is harder than reviewing human-written code because context must be rebuilt from scratch, making readability and good coding practices more important than ever. The author argues that established principles like DDD and Clean Architecture remain relevant — not for machines, but for the humans who must own what they shi

Moonshot AI dropped the full weights for Kimi K3 this week, and the reaction has been loud. The model is a 2.8-trillion-parameter Mixture-of-Experts that activates only 104B parameters per request, compressed to ~1.56TB in MXFP4. Benchmarks put it level with Anthropic’s and OpenAI’s current closed flagships, at a fraction of the API cost. The integration story moved fast. Kimi K3 is now available in GitHub Copilot across all plan tiers (Pro through Enterprise), hosted on Fireworks AI, and access

A critical vulnerability in ColdCard Bitcoin hardware wallets led to the theft of over 1,600 Bitcoin (~$115 million) from more than 7,000 wallets. The root cause was a subtle firmware bug: ColdCard's MicroPython-based firmware had two random number generators with the same function name. A flag intended to disable MicroPython's weak built-in RNG was set to zero rather than being undefined, causing an 'if not defined' check to pass and the weak RNG to be used instead of ColdCard's secure one. On

Next.js 16.3 ships two major improvements: automatic performance gains requiring no code changes, and opt-in Instant Navigations. Default upgrades include up to 90% less dev server memory via Turbopack memory eviction, faster repeat builds with Turbopack FileSystem Cache (up to 5.5x speedup), 22% more SSR throughput via native Node.js streams, and TypeScript 7 support for 10x faster type checking. The headline feature is Instant Navigations, enabled via two config flags, which brings client-side

Stephen Wolfram shares a deeply personal tribute to his wife, Elise Cawley (1961–2026), a brilliant pure mathematician who died suddenly following heart surgery. He recounts their 36 years together, her academic achievements — including a PhD on smooth Markov partitions, a tenure-track position at Stony Brook, and work at the University of Chicago — her passion for mathematical elegance, her role as a mother of four, her contributions to interior design and architecture, and her philosophical vi

A critique of overengineering in software development, arguing that senior developers and CTOs often build overly complex architectures for problems they don't yet have. Uses humor to illustrate the 'architect's ego' trap, then provides a quick overview of seven architecture patterns (layered, microservices, event-driven, micro kernel, serverless, space-based, hexagonal) with guidance on when to use or avoid each. The core message: simplicity enables scale, not the other way around.

Tabularis v0.18.0 is a substantial release focused on database administration and usability. Key additions include a full Users & Privileges management UI for MySQL and MariaDB (list, create, edit grants), a rebuilt connection test modal that classifies SSH/tunnel/DB errors step-by-step with a copyable diagnostics report, and an all-databases mode that auto-discovers databases without requiring upfront selection. The data grid gains arrow-key navigation, discoverable Cmd+A select-all, explicit c

A reflection on the gap between memorizing Kubernetes architecture and truly understanding it through hands-on experience. The author recounts a production CrashLoopBackOff incident where diagram-level knowledge proved insufficient, and a senior engineer's instinctive use of `kubectl logs --previous` revealed how real understanding is built from debugging experience, not documentation. Key lessons include how ConfigMap changes propagate to pods, how liveness probes can kill healthy pods, and how