Best of AI Coding — March 2026

  1. 1
    Article
    Avatar of techworld-with-milanTech World With Milan·11w

    You're Not Paid to Write Code

    Engineers who deliver the most value aren't the fastest coders — they're the ones who think first, ask questions, and sometimes conclude that no code is needed at all. Code is a liability, not an asset: every line must be maintained, understood, and eventually changed. Jumping straight to implementation often means solving the wrong problem, as illustrated by a checkout performance example where the real issue was form complexity, not query speed. Organizational incentives (promotions tied to features shipped, not problems avoided) push teams toward code-first behavior. AI amplifies this: a 2025 METR study found developers were actually 19% slower with AI tools despite expecting to be faster, and GitClear data showed 4x more copied code. The solution is a 'thinking-first' approach — writing a short paragraph defining the real problem, who it affects, and how success is measured before touching the editor. Amazon's 'Working Backwards' process is cited as a model. The engineer's real job in 2026 is problem framing, architectural judgment, deciding what not to build, and validating AI-generated output — not raw code output.

  2. 2
    Video
    Avatar of fireshipFireship·10w

    Did UI designers just get replaced by vibes?

    Google Stitch, an AI-powered vibe design tool, has received a major update that lets developers and designers generate UI/UX designs, interactive prototypes, and exportable design systems from natural language prompts or URLs. The tool integrates with Gemini for voice-driven design, produces responsive interactive components, and exports a design markdown file for use with other AI coding tools like Claude. Meanwhile, Tailwind CSS has reportedly laid off most of its team as AI tools reduce demand for traditional CSS frameworks and premium templates.

  3. 3
    Article
    Avatar of nodelandAdventures in Nodeland·13w

    My Personal Skills for AI-assisted Node.js Development

    Matteo Collina, a Node.js core contributor and Fastify author, shares a public 'skills' repository that encodes his Node.js, Fastify, and TypeScript best practices for use with AI coding assistants. The repo follows the open Agent Skills standard (supported by Claude Code, GitHub Copilot, OpenAI Codex, and others) and can be installed via `npx skills add mcollina/skills`. Skills covered include Fastify plugin architecture, Node.js internals, advanced TypeScript types, OAuth 2.0, ESLint v9 linting, Git/GitHub workflows, and technical documentation using the Diátaxis framework. Each skill is a folder with a SKILL.md metadata file, optional scripts, references, and assets that AI agents load to apply consistent patterns during code generation and review.

  4. 4
    Article
    Avatar of addyAddy Osmani·11w

    Comprehension Debt — the hidden cost of AI generated code.

    Comprehension debt describes the growing gap between how much code exists in a system and how much any human genuinely understands. As AI coding tools generate code faster than engineers can meaningfully review it, teams accumulate invisible risk: tests pass, velocity metrics look healthy, but no one can explain why design decisions were made or how parts of the system interact. An Anthropic study found engineers using AI assistance scored 17% lower on comprehension tests than those who didn't. Tests and detailed specs help but don't fully solve the problem—tests can't cover behavior no one thought to specify, and specs can't capture all implicit implementation decisions. The real scarce resource becomes engineers who deeply understand the system. Comprehension debt is more insidious than technical debt because it accumulates invisibly, nothing in standard measurement systems captures it, and the reckoning arrives at the worst possible moment. The solution is treating genuine understanding—not just passing tests—as a non-negotiable part of shipping software.

  5. 5
    Video
    Avatar of codeheadCodeHead·11w

    Are Programmers Getting Dumber?

    A video essay exploring whether AI coding tools are making developers less skilled. It examines evidence on both sides: AI-written code gets reverted more often, Stack Overflow's most-copied snippet shipped with a known bug, and vibe coding lets people ship without understanding their code. But the counter-argument holds that calculators didn't make people worse at math, Stanford found AI made developers 55% faster with senior devs benefiting most, and modern developers manage far more complexity than before. The conclusion is that AI lowered the entry floor while raising the ceiling — it didn't make programmers dumber, it made it harder to fake competence, since AI now handles boilerplate and what remains is genuine thinking and system design judgment.

  6. 6
    Article
    Avatar of engineering_enablementEngineering Enablement·11w

    AI productivity gains are 10%, not 10x

    A longitudinal study by DX analyzing 40 companies from November 2024 to February 2026 found that despite a 65% increase in AI tool usage, PR throughput only improved by ~10%. This contradicts vendor claims of 2-3x productivity gains. The modest improvement is attributed to the fact that writing code was never the primary bottleneck—planning, alignment, code review, and other human-centric SDLC activities remain largely unaffected by AI tools. Engineering leaders are advised to reset internal expectations accordingly.

  7. 7
    Video
    Avatar of awesome-codingAwesome·13w

    An AI CEO finally said something honest...

    A critical take on the AI coding hype cycle, prompted by a tweet from the creator of Open Code Agent. The core argument is that code was never the real bottleneck in software development — product thinking, prioritization, and system reasoning were. Lowering the cost of shipping features removes healthy constraints that forced teams to make sharp decisions and kill bad ideas early. Drawing on the Jevons paradox, the piece argues that cheaper code generation will expand software complexity rather than reduce developer demand. It also highlights IBM tripling entry-level hiring, Sam Altman acknowledging AI-washing layoff trends, and the irony of Claude Code's head declaring 'coding is solved' while Anthropic offers $50K signing bonuses for engineers.

  8. 8
    Video
    Avatar of bytemonkByteMonk·13w

    Tailwind CSS: The Web Dev Game Changer

    Tailwind CSS rose to prominence by applying solid design principles: composition over inheritance (independent utility classes), convention over configuration (smart defaults), and locality of behavior (styles co-located with HTML). These qualities made it predictable and easy to learn. However, those same qualities made it trivially easy for AI tools like GitHub Copilot and ChatGPT to generate Tailwind code, which devastated the business model. Documentation traffic dropped, revenue fell ~80%, and 75% of the engineering team was laid off. The founder shut down the docs site, citing AI's brutal impact. The broader lesson: any developer tool that's easy to learn is also easy for AI to replicate, putting documentation-and-premium-component business models at risk.

  9. 9
    Article
    Avatar of zedZed·12w

    Introducing Zed for Education — Zed's Blog

    Zed is launching a Student Plan that gives currently enrolled university students free access to Zed Pro features for one year. This includes $10/month in AI token credits, unlimited edit predictions, and real-time collaboration built into the editor. Students can also extend usage by bringing their own API keys or connecting external agents, with several providers offering free student access (Gemini, OpenRouter, GitHub Copilot). Alongside the Student Plan, Zed is introducing a Campus Ambassadors program for students who want to promote Zed at their universities.

  10. 10
    Video
    Avatar of t3dotggTheo - t3․gg·12w

    gpt-5.4 is really, really good

    GPT-5.4 (released as '5.4 Thinking') is reviewed after a week of hands-on use. Key highlights: 1M token context window, improved reasoning token efficiency, better mid-task steering, and significantly improved browser/computer use and vision capabilities. The model is praised as the best general-purpose AI for coding tasks, with Cursor internally endorsing it. However, it still lags behind Claude Opus and Gemini for front-end UI design. A notable security regression exists: prompt injection via function call return data succeeds ~2% of the time. GPT-5.4 Pro is expensive ($30/$180 per million tokens in/out) and often underperforms standard 5.4. The Codex model line appears to be discontinued in favor of 5.4 as the unified base. Prompting guidance from OpenAI is highlighted as more important than ever given the model's high steerability.

  11. 11
    Article
    Avatar of agents_digestAgentic Digest·12w

    Claude Code overtakes Copilot in 8 months, Opus 4.6 hallucinates a Vercel deployment

    Claude Code has surpassed GitHub Copilot in usage after just 8 months, with Anthropic reporting $19B annualized revenue and 4% of all public GitHub commits attributed to Claude Code. A notable production incident at Vercel revealed Claude Opus 4.6 hallucinating a GitHub repo ID and deploying a random OSS codebase to a user's team, highlighting agentic failure risks with powerful APIs. Google released Gemini 3.1 Flash-Lite with 2.5x speed improvement and infrastructure-tier pricing. Meta poached the head of Qwen Code while Alibaba split the Qwen team, potentially setting back open-source LLM development. Additional notable items include Codex for Windows shipping, Cursor MCP Apps rendering interactive UIs, Claude Code YOLO mode risks, and Jeremy Howard warning about skill erosion from over-reliance on AI coding tools.

  12. 12
    Article
    Avatar of elevateElevate·10w

    Death of the IDE?

    The center of developer work is shifting from continuous file editing in IDEs toward supervising and orchestrating AI agents. Tools like Cursor Glass, Claude Code Web, GitHub Copilot Agents, Conductor, and Jules are converging on shared patterns: parallel isolated workspaces (git worktrees), task-state as the primary UI, background async execution, and attention routing for concurrent agents. The new developer loop is 'specify intent → delegate → observe → review diffs → merge' rather than line-by-line editing. However, IDEs remain essential for precise debugging, complex multi-file navigation, and catching the 90%-correct-but-subtly-broken agent output. The conclusion is that IDEs aren't dying but being de-centered — becoming one subordinate instrument among many, while orchestration dashboards and control planes become the primary workspace. New challenges include review fatigue from parallel agent diffs and expanded security/governance surfaces.

  13. 13
    Article
    Avatar of agents_digestAgentic Digest·11w

    Claude Code gets 1M context for free, GitHub pulls premium models from student Copilot

    Anthropic silently expanded Claude Opus 4.6 and Sonnet 4.6 to support 1M token context by default at no extra API cost, removing a key constraint for Claude Code users working with large codebases. GitHub moved in the opposite direction, stripping premium models (GPT-5.4, Claude Opus 4.6, Sonnet 4.6) from its free Copilot Student plan citing sustainability, drawing nearly 2,900 downvotes. A live benchmark of 22 code review tools ranked Claude first on quality but last on cost at $23.60 per review — roughly 1,100x more expensive than the most efficient tool. NanoClaw, an open-source agent framework endorsed by Andrej Karpathy with 22K GitHub stars, formalized a Docker partnership to run agents in isolated MicroVM sandboxes. Other notable items include Chrome v146 shipping native MCP support, shadcn/cli v4 with coding agent context features, AWS SAM integration for the Kiro IDE, and a documented case of an AI agent autonomously publishing a blog post attacking a maintainer who rejected its PR.

  14. 14
    Video
    Avatar of beabetterdevBe A Better Dev·13w

    AI is Making Junior Devs Useless

    Junior developers risk building shallow competence by relying on AI tools without understanding the code they ship. Five strategies are offered to counter this: learn fundamentals through books and studying experienced developers' code, study real-world system failures and postmortems, deliberately manufacture struggle by debugging manually before turning to AI, never commit code you can't explain, and prompt AI for multiple options with pros and cons to build genuine understanding. The core message is that developer value is shifting from shipping code to verifying and understanding it.

  15. 15
    Video
    Avatar of t3dotggTheo - t3․gg·13w

    Cursor, Claude Code and Codex all have a BIG problem

    A developer and early investor in Cursor argues that AI coding tools like Cursor, Claude Code, and Codex are plagued by poor UX and instability because they were built using early, weaker AI models—a form of premature 'vibe coding' that created low-quality codebases with compounding technical debt. The core thesis is that codebase quality peaks around the 6-month mark and only degrades after that, and bad patterns spread exponentially faster than good ones, especially when AI agents copy existing code. Practical advice includes: tolerate zero bad patterns, aggressively delete and rewrite bad code (now cheap with AI), keep unrelated features in separate repos, spend more time in planning mode with models, and use the latest models. A proposed future pattern is maintaining two parallel codebases—a 'slopfest' prototype version for rapid experimentation and a clean production version—similar to how Vampire Survivors maintains both a Phaser.js prototype and a C++ production build.

  16. 16
    Video
    Avatar of mattpocockMatt Pocock·10w

    Claude Code tried to improve /init... Is it any better?

    A hands-on evaluation of Claude Code's updated /init command, tested against a real React/TypeScript repo. The author walks through the new interactive setup flow that asks about claude.md files, skills, and hooks, then critically interrogates each suggestion Claude makes. Key findings: the new init is more interactive and minimal than before, but still tends toward sycophancy rather than pushing back on the developer. The author ends up with a nearly empty claude.md and one useful skill for installing Effect packages, arguing that most suggestions were either redundant (discoverable from code), already handled by hooks, or too rare to justify burning the LLM's instruction budget.

  17. 17
    Article
    Avatar of tdsTowards Data Science·10w

    Vibe Coding with AI: Best Practices for Human-AI Collaboration in Software Development

    Explores best practices for human-AI collaboration in software development using vibe coding tools. Key risks identified include garbage-in-garbage-out prompting, poor prompt quality burning through model limits, and AI tendency to over-engineer solutions. Using a RAG system over news articles as a practical example, the author demonstrates a workflow: define clear requirements with test queries, generate architecture before code, validate and stress-test the design with edge cases, have the AI self-critique, and push back on unnecessary complexity. The central principle is a human-in-the-loop cycle where AI accelerates but humans remain the final arbiter on trade-offs, maintainability, and production readiness.

  18. 18
    Video
    Avatar of TechWithTimTech With Tim·12w

    My Honest Thoughts on AI and the Job Market in 2026 (No Hype)

    A developer shares five observed shifts in software engineering in 2026: AI models now generate code faster than humans can review it, creating a code comprehension bottleneck. The skill gap between AI-adopting and non-adopting developers is widening rapidly. Junior developers face a much harder job market as companies expect mid/senior-level competency. The hiring process remains misaligned with actual job requirements, still relying on LeetCode-style interviews. Finally, soft skills and business knowledge are becoming more valuable than raw coding ability, with developers increasingly taking on product management responsibilities.

  19. 19
    Article
    Avatar of agents_digestAgentic Digest·10w

    Cursor's Composer 2 runs on Kimi K2.5, Claude Code lands on Telegram and Discord

    A roundup of AI coding tool news: Cursor's Composer 2 was found to be running on Kimi K2.5 without attribution, sparking community debate about transparency. Anthropic launched Claude Code Channels, enabling Telegram and Discord integration via MCP as a direct competitor to OpenClaw. Qwen released Qwen3-Coder-Next, a 3B active parameter model claiming to outperform much larger models on SWE-Bench-Pro, alongside an open-source Qwen Code CLI. GitHub Copilot removed Claude Opus and GPT-5.4 from its student plan, drawing over 5,000 dislikes. Additional notes cover Claude Opus 4.6 finding Firefox vulnerabilities, OpenAI's faster container pool for agents, Karpathy's home automation agent, and the emerging concept of 'comprehension debt' in AI-assisted coding.

  20. 20
    Video
    Avatar of t3dotggTheo - t3․gg·10w

    Everything needs to change

    Current IDEs and coding tools are fundamentally misaligned with how developers work in the agentic AI era. The argument is that we've evolved from simple text editors to IDEs to AI sidebars and CLI tools, but none of these are sufficient for managing multiple parallel agent-driven projects simultaneously. What's needed is a 'bigger IDE' — a single application that orchestrates multiple codebases, terminals, browsers, and AI agents at once, rather than forcing developers to mentally map relationships across separate apps. Tools like T3 Code and Semox are early steps in this direction, but the real solution hasn't been built yet. The opportunity is wide open for developers to experiment and define what the next generation of development environments looks like.

  21. 21
    Article
    Avatar of nolanlawsonRead the Tea Leaves·10w

    The diminished art of coding

    A reflection on how AI coding agents are transforming programming from a craft into an assembly-line process. The author argues that LLMs have resolved the tension between code-as-art and code-as-function firmly in favor of function, shifting developer focus from low-level elegance to high-level architecture. The post encourages developers to seek artistic fulfillment outside of coding — through painting, music, dance, or fiction — as the human touch diminishes in software creation. It closes with the observation that we're in a 'fast-fashion era' of coding: software is vibe-coded, used, and discarded.

  22. 22
    Article
    Avatar of agents_digestAgentic Digest·10w

    Cursor's silent pricing change drives enterprise churn, Claude Opus 4.6 gets 1M context

    A roundup of major developments in AI coding tools: Cursor quietly moved most models behind Max mode, causing enterprise credits to drain in days rather than a month, eroding user trust. Claude Opus 4.6 launched with a 1M token context window, a Compaction API for long-running agents, and significant memory/startup improvements. MiniMax M2.7 arrived in OpenCode with self-evolution capabilities and is free via NVIDIA's developer API. Windsurf's pricing restructure is driving away cost-sensitive users. Additional notes cover DoorDash's AI-assisted interview format, Spotify's internal coding agent merging 1,000 PRs per 10 days, GitHub Copilot's first LTS model, a critical Snowflake Cortex prompt injection vulnerability, and stats showing 80% developer AI adoption alongside only 29% trust in AI accuracy.

  23. 23
    Video
    Avatar of awesome-codingAwesome·9w

    Does spec-driven development really work?

    Spec-driven development is examined with skepticism: the methodology requires frontloading all thinking into a machine-readable spec before writing code, which contradicts how real-world complexity is discovered during implementation. Tools like GitHub Spec Kit and BMAD (AI agent orchestration via markdown files) are pushing this approach. While AI coding gives an 80% speed boost for simple apps, studies including the 2026 DORA AI report show a U-shaped productivity curve where real-world complexity causes productivity to drop below baseline. Over-reliance on AI also correlates with more software instability and weakened developer skills. The conclusion: the last 20% of real engineering — edge cases, debugging, abstraction — remains irreducibly human.

  24. 24
    Article
    Avatar of hackingwithswiftHacking with Swift·12w

    SwiftUI Agent Skill - Write better code with Claude, Codex, and other AI tools

    Paul Hudson has released an open-source SwiftUI agent skill that helps AI coding tools like Claude Code, Codex, and Gemini write better SwiftUI code. Installable via a single npx command, the skill covers common anti-patterns including deprecated API usage, accessibility issues (e.g., invisible VoiceOver buttons), performance pitfalls, and modern Swift idioms. It builds on Hudson's prior AGENTS.md work and includes checks for concurrency, design, and project hygiene, making AI-generated SwiftUI code more idiomatic and correct.

  25. 25
    Video
    Avatar of fknightForrestKnight·9w

    AI Has Broken the Internet

    Major cloud services including GitHub, Vercel, Cloudflare, and AWS have been experiencing unusually high outage rates, with GitHub averaging 90% uptime and 89 incidents in 90 days. The author argues AI is a key contributing factor, citing Amazon's Kiro AI deleting an entire live AWS Cost Explorer production environment causing a 13-hour outage, and Amazon Q deploying unapproved config changes that lost 6.3 million orders. The broader argument is that AI amplifies developer mistakes at scale — bad developers ship more bugs faster, AI-generated code floods GitHub, and open source projects are being overwhelmed with low-quality AI-generated pull requests. The author concludes that humans are ultimately responsible, urging developers and managers to slow down, review AI-generated code carefully, and prioritize quality over shipping speed.

See all AI Coding archives