Best of AI Coding — April 2026

1
Article
Xavier Womack·6w
So AI is making me Lazy...
A developer reflects on how AI code assistants can quietly erode core engineering skills — reading documentation, debugging from first principles, and architecting from scratch. The post warns against over-reliance on AI-generated code that passes tests but hides subtle bugs, and offers practical habits to stay sharp: reading every line of AI output, building something from scratch monthly without AI help, owning architecture decisions, and rubber-ducking generated code. The author uses a calculator analogy to distinguish between using AI as a productivity multiplier versus a crutch that atrophies foundational understanding.
584
94
2
Article
Anubhav Bhatt·5w
DESIGN.md - UI design systems format shared with AI
Google's Stitch has introduced DESIGN.md, an open-source file format designed to standardize how UI design systems are shared with AI tools. It combines machine-readable design tokens (colors, typography, spacing) with human-readable context explaining design intent, enabling AI to generate consistent, brand-aligned interfaces without repeated manual explanation. The format includes validation tools like accessibility checks and integrates with workflows such as Tailwind and CI pipelines, aiming to become a universal design contract that reduces friction and improves designer-developer collaboration.
580
62
3
Article
sean goedecke·5w
Software engineering may no longer be a lifetime career
A thought-provoking argument that the traditional software engineering career path may be fundamentally changing due to AI. Even if using AI tools causes skill atrophy over time, engineers may still be obligated to use them to remain competitive — just as construction workers must lift heavy objects despite long-term physical wear. The analogy to professional athletes with a finite career window is drawn: software engineers may need to plan for a career with a limited lifespan rather than assuming lifelong skill accumulation.
318
85
4
Article
freeCodeCamp·6w
The New Definition of Software Engineering in the Age of AI
AI is not replacing software engineers wholesale — it's automating routine, execution-level coding tasks. The shift demands developers move from effort-based to impact-based engineering: understanding system architecture, applying clean code principles, debugging complex distributed systems, and taking ownership of outcomes. A five-step roadmap is outlined: strengthen CS fundamentals, build real-world systems with failure handling, master debugging, use AI as a tool rather than a crutch, and establish proof of work through public building and open-source contributions. The core argument is that source code is now a byproduct of thinking, not the primary output.
244
19
5
Video
CodeHead·6w
Should You... Keep Coding?
A reflective piece questioning whether developers should keep coding amid AI disruption and tech layoffs. It argues that many so-called AI layoffs are actually cover for pandemic-era overhiring mistakes, and that vibe coding has a real understanding gap — non-technical users can't debug what they don't understand. The piece emphasizes that coding teaches problem-solving and structured thinking that remains valuable regardless of market shifts. Ends with a sponsored recommendation for Coursera courses in Python, Unix, and software architecture.
158
36
6
Article
swizec.com·5w
AI now writes 97% of my code. Here's what I learned
A senior engineer at a near-billion-dollar company shares that AI (via Cursor) now writes 97% of their production code, including high-stakes systems like an invoicing platform. The key workflow: use Cursor background agents via Slack, don't watch them work, provide clear context and desired outcomes in prompts, then iterate with code review just like you would with a human engineer. The author dismisses elaborate prompt engineering tricks (Agents.md, plan mode, etc.) as distractions, arguing the fundamentals — speak clearly, give context, iterate — are what actually matter. A cautionary note: long-running agents that make early mistakes can produce thousands of lines of broken code, echoing the failures of waterfall development.
174
33
7
Article
Kilo Blog·6w
Thank you, Roo! We’ll take it from here.
Roo Code is shutting down and archiving its repo on May 15th after 3 million installs. Kilo, which started as a fork of Roo Code, announces it is continuing full-speed development of its VS Code extension. The extension was recently rebuilt from the ground up on the OpenCode server, gaining parallel execution, subagent delegation, an Agent Manager, inline diff review, and cross-platform sessions. Kilo invites former Roo contributors and users to join their open-source project.
99
4
8
Article
Lobsters·8w
I used AI. It worked. I hated it.
A developer with strong anti-genAI views shares their experience using Claude Code to build a certificate generation tool for a learning platform migration. The project succeeded — the Rust/Svelte app is in production — but the process felt miserable. Key observations: TDD with plan-mode kept the model on track, Rust's compile-time safety helped catch hallucinations, a security audit pass found real vulnerabilities (path traversal, timing side-channel in Argon2), and the tool delivered features the author wouldn't have built alone. Despite the functional success, the author warns about the 'human in the loop' problem — the process actively encourages disengagement — and reflects on cognitive dependency, skill atrophy for junior developers, IP theft concerns, and the broader societal harms. The conclusion is nuanced: it works in this narrow domain, the harms still outweigh benefits at scale, but condemning individuals for using it is counterproductive.
99
18
9
Article
FireUp·5w
and it might replace the way we share UI Design forever : fireup.pro
Google has unveiled Stitch, an experimental AI design tool from Google Labs that generates UI layouts and frontend code from text descriptions. The more significant announcement is DESIGN.md, a machine-readable design specification format that replaces static files like Figma exports or PDFs. DESIGN.md enables AI tools to generate consistent UI from shared rules, allows developers to integrate design intent directly into code pipelines, and makes design version-controllable like software. The format represents a shift from 'design as a file' to 'design as a system-readable spec', potentially transforming how design and development teams collaborate.
119
12
10
Article
Hacker News·8w
The Claude Code Leak
The accidental leak of Claude Code's source code reveals that the code itself is reportedly low quality, yet the product is beloved. This prompts reflection on five key observations: code quality doesn't determine product success; what matters is what code does, not how it's written; product-market fit trumps implementation details; the copyright situation is ironic given Anthropic's own AI training arguments; and ultimately the leak won't matter because Claude Code's value lies in the seamless integration of model and harness, not the underlying source code. Open-source alternatives like Codex and Gemini CLI haven't captured Claude Code's mindshare, reinforcing that a complete, well-integrated service is what users pay for.
67
9
11
Video
Web Dev Cody·7w
AI killed the joy of coding
A veteran web developer with 13 years of experience shares his personal struggle with the loss of joy in coding since AI tools like Claude Code and Cursor became dominant. He reflects on how agentic coding has replaced the satisfaction of hands-on problem solving, discusses the anxiety of managing multiple AI agents, and questions the future of his career and YouTube channel. He also touches on the difficult job market, the rise of vibe coding by non-developers, and where experienced developers might still have an edge — in deep technical knowledge or business/marketing skills.
66
16
12
Article
Agentic Digest·7w
Claude Opus 4.6 gets quietly nerfed, Grok 4.20 tops BridgeBench
Claude Opus 4.6's thinking budget was quietly cut 67% (from 100 to 25), causing noticeable drops in reasoning quality for subscribers. xAI's Grok 4.20 now leads BridgeBench over GPT-5.4 and Opus 4.6. Anthropic's unreleased Mythos model — capable of autonomously discovering zero-day vulnerabilities and scoring 93.9% on SWE-bench — is restricted to a consortium of AWS, Apple, Google, and Microsoft via Project Glasswing. Vercel open-sourced Open Agents, a reference platform for cloud-based coding agents. Additional updates include Cursor 3 agent splitting, GitHub Copilot data residency and merge conflict fixes, a Microsoft MEMENTO research finding on KV cache persistence, Cloudflare Sandboxes GA, and a Stanford study showing frontier models score 70–80% on vision benchmarks even without images.
64
5
13
Video
Web Dev Simplified·7w
Can I Build This UI In 10 Minutes
A developer attempts to recreate a UI design using only CSS in 10 minutes as part of a timed challenge on CSSBattle-style platform. The walkthrough covers planning the layout using flexbox, applying CSS variables for colors, styling score rows with dotted borders, and fine-tuning typography and spacing. After achieving a 73% match manually, the author tests the same challenge with AI, which scores only 61% and produces quirky artifacts like unnecessary linear gradients, hardcoded pixel values, and ignored CSS variable instructions — concluding that AI is a useful starting point but not a reliable replacement for hands-on CSS skills.
62
1
14
Video
Fireship·8w
Cursor ditches VS Code, but not everyone is happy...
Cursor 3.0 marks a major shift from its VS Code fork origins, now rewritten from scratch in Rust and TypeScript with a focus on orchestrating swarms of AI agents across multiple repos, machines, and cloud environments simultaneously. The release also introduced Composer 2, an in-house coding model that sparked controversy after it was revealed to be based on Moonshot's Kimi K2 model — a fact Cursor initially obscured, later apologizing for the lack of transparency. The new interface de-emphasizes manual coding in favor of agent management, featuring parallel agent monitoring, built-in browser, design mode, and remote SSH support. Not everyone is enthusiastic about this direction, with some critics comparing it too closely to OpenAI Codex.
58
1
15
Video
Philipp Lackner·6w
3 Theoretical Limits of AI - These Things Can't Be Fixed
A critical look at three fundamental, unfixable limitations of current LLM-based AI: (1) the learning ceiling problem — LLMs can't exceed the collective intelligence of their training data, especially as AI-generated content pollutes future training sets; (2) hallucination as an architectural inevitability — the same mechanism that enables creativity also produces confident incorrect outputs, and these can't be separated; (3) the frame problem — LLMs operate strictly within the context given to them and lack the ability to reframe a problem the way an experienced developer would. The author argues the truth lies between AI replacing developers and AI being useless, and that developers who understand these limits and use AI skillfully will gain a real productivity edge.
52
1
16
Article
freeCodeCamp·8w
AI Tools for Developers –
A freeCodeCamp course covering AI tools for developers, including GitHub Copilot, Anthropic's Claude Code, and Gemini CLI for AI pair programming and agentic terminal workflows. Also covers OpenClaw for locally hosted open-source AI automation and CodeRabbit for AI-driven pull request analysis. The course is 1.5 hours and available on YouTube.
43
2
17
Article
Collections·8w
Qwen 3.6 Plus: Alibaba's agentic coding model with 1M context, now free via Qwen Code and OpenRouter
Alibaba has released Qwen 3.6 Plus, a closed-weights model targeting agentic coding and repository-level tasks with a 1 million token context window and multimodal input. It supports improved tool-calling and long-horizon planning for frontend and repo-scale coding. Access is available for free via Qwen Code CLI (1,000 requests/day via OAuth), OpenRouter, Qwen Chat, and Vercel AI Gateway under the identifier `alibaba/qwen3.6-plus`. The large context window is positioned as the key differentiator for loading entire codebases into context.
36
3
18
Video
Awesome·8w
They want mediocre developers...
A critical take on 'comprehension debt' — the growing gap between the volume of AI-generated code in a system and how much of it developers actually understand. Drawing on a study showing AI-assisted developers performed worse at debugging, and the 2024 DORA report showing higher throughput but increased delivery instability, the argument is that AI coding tools are accelerating code production faster than teams can meaningfully review or comprehend it. The concern extends to a long-term lock-in scenario: AI labs are subsidizing low prices during a customer acquisition phase, and as developers lose critical thinking and debugging skills through over-delegation, they become dependent on tools whose prices could rise dramatically. The post also critiques the incentive structure of AI labs profiting from token consumption regardless of code quality.
32
3
19
Article
Claude·7w
Redesigning Claude Code on desktop for parallel agents
Anthropic has redesigned the Claude Code desktop app to support parallel agentic workflows. Key additions include a new sidebar for managing multiple concurrent sessions across repos, drag-and-drop workspace layout, an integrated terminal and file editor, a faster diff viewer, expanded preview support for HTML and PDFs, and SSH support extended to Mac. Three view modes (Verbose, Normal, Summary) let users control how much detail they see. The app now streams responses in real time and has parity with CLI plugins. Available to Pro, Max, Team, and Enterprise plan users and via the Claude API.
28
4
20
Video
Philipp Lackner·7w
Chill out.
A measured take on AI hype in software development, arguing that the truth lies between extreme skepticism and uncritical belief in AI CEO predictions. Developers are encouraged to adopt one AI coding tool (Claude Code is personally recommended) and master it rather than chasing every new model or benchmark. AI is genuinely changing software development but is far from autonomously replacing developers, especially in complex enterprise contexts. The advice: engage with agentic coding workflows, but don't let FOMO drive anxiety-driven tool-hopping.
28
5
21
Article
Faun·4w
Qwen3.6–35B-A3B: The Most Practical Open-Source AI Model Yet?
Qwen3.6-35B-A3B is a Mixture-of-Experts open-source model with 35B total parameters but only ~3B active per request, making it highly efficient. It features a 262K context window (extendable to 1M with YaRN), multimodal support (text, image, video), and an Apache 2.0 license. The model is designed for agentic coding workflows, achieving top scores on SWE-bench Verified (73.4), Terminal-Bench 2.0 (51.5), and strong STEM reasoning benchmarks. Key architectural innovations include Gated DeltaNet linear attention and Grouped Query Attention (GQA). It supports a switchable thinking/non-thinking mode and a new thinking preservation feature that reuses reasoning across conversation turns. Deployment is supported via vLLM, SGLang, KTransformers, and Hugging Face.
76
6
22
Article
iO tech_hub·5w
Coding Is Being Commoditized. Engineering Is Not.
Agentic coding tools are making code generation cheap and fast, but this doesn't commoditize engineering itself. The real value shifts to judgment, systems thinking, and accountability — skills that experienced engineers have built through years of debugging, architecture decisions, and production failures. The post argues that engineers in the 'sweet spot generation' are well-positioned to amplify their output using agents while protecting system integrity. It outlines what 'agentic engineering' looks like in practice: framing work for agents, maintaining architectural invariants, reviewing generated code with full accountability, designing safe workflows, and understanding blast radius. Practical advice is given for both senior engineers (treat agents like strong juniors) and early-career developers (build fundamentals first, use agents to learn rather than shortcut).
26
2
23
Article
Zed·7w
Introducing Zed's Agent Stats — Zed's Blog
Zed has launched Agent Stats, a public weekly dashboard showing AI agent adoption and performance metrics inside the Zed editor. The data covers session counts, turn volumes, and response time distributions (p10/p50/p90) across agents. Key findings: Claude Sonnet's p90 latency rose 44% over three weeks (294s to 425s), Claude Code leads on depth with 10.6 turns per session vs Zed's native agent at 5.5, and 10 distinct GPT-5 variants are currently active. The data comes from opted-in users only and covers 2 million sessions and 15.4 million turns over 90 days. Any agent implementing the open Agent Client Protocol can appear in the stats.
23
2
24
Article
The Miners·7w
Stop Putting Best Practices in Skills
A data-driven investigation into why best practices should live in CLAUDE.md rather than Claude Code skills. The author ran 51 multi-turn evals across 4 configurations (Superpowers, plain skills, CLAUDE.md, CLAUDE.md+hint) and found that plain skills are only invoked 6% of the time in multi-turn sessions, while CLAUDE.md guidelines are always in context. The key insight: skills and CLAUDE.md are both just prompts — the difference is reliability of delivery. Superpowers works not because of skills but because its SessionStart hook front-loads instructions, achieving 66% invocation. The recommendation is clear: put coding standards, TDD rules, and debugging protocols in CLAUDE.md (100% presence, no activation gap), and reserve skills for on-demand procedural recipes like scaffolding or migrations.
22
25
Video
Nick Chapsas·7w
This is Why AI Could Replace Programmers
A developer argues strongly against the notion that programmers no longer need deep coding knowledge because AI tools can write code for them. While acknowledging that AI coding tools like Claude Code, Codex, and Cursor have changed how developers work, the author contends that domain expertise remains critical for writing secure, production-ready code. Key points include: AI-generated code without expert oversight leads to security vulnerabilities; larger LLM context windows don't necessarily produce smarter code; developers who rely solely on AI without understanding the underlying language are competing with raw compute rather than offering unique value. The author advocates for learning to use AI tools as a speed multiplier while maintaining deep technical knowledge, and warns that developers who abandon learning will be replaced.
22
7

See all AI Coding archives