#vibe-coding

Is Successful Agentic Coding a Delusion?

Avdi Grimm so sánh sự hoài nghi trước đây về XP (như TDD) với làn sóng hoài nghi hiện nay về agentic coding. Ông thừa nhận những lo ngại về môi trường và đạo đức của AI nhưng khẳng định LLM-assisted coding thực sự hiệu quả khi áp dụng nghiêm túc (review, test, lint, cộng tác chặt chẽ) thay vì "vibe coding". Ông cảnh báo các nhà phát triển giỏi từ chối công nghệ này có thể tự đẩy mình ra ngoài cuộc chơi.

Những lập trình viên đã từng thắc mắc về hiệu quả thực tế của AI hỗ trợ mã hóa nên đọc để tránh bị lờ đi những cơ hội hiệu quả khi áp dụng phương pháp nghiêm ngặt, thay vì chỉ là những lo ngại về hype hay rủi ro.

Python Is Not Enough: Why Pythonistas Love Rust (Podcast)

A podcast episode recap covering why Python developers are adopting Rust, featuring discussion of Rust tooling (Ruff, uv, Polars, Pydantic core), how Rust's strictness benefits AI agent guardrails, and a skeptical take on vibe coding. The author argues real AI productivity gains are closer to 1.2-2x rather than 10x, warns about rubber-stamping AI-generated code, and emphasizes that deep engineering fundamentals outlast hype cycles. The post also promotes a 6-week Python-to-Rust cohort building a JSON parser with PyO3 bindings.

Don't Be Careful, Be Competent

A payments engineer argues against fully delegating code to LLMs, making the case that writing code yourself is the highest form of thinking — not low-level work. Drawing on Peter Thiel's 'Zero to One' and Brooks' 'Mythical Man Month', the author contends that AI-native companies that will succeed are those where engineers still write code by hand, maintaining conceptual integrity and a unified product vision. Vibe coding is framed as a Faustian bargain: convenient but corrosive to deep understanding, flow state, and the nuanced knowledge that differentiates great engineers. The recommended approach is to use LLMs for brainstorming and review, not delegation.

Systematic AI Coding: My Takeaways from the Eclipse Foundation Workshop in Brussels

A developer's takeaways from the Eclipse Foundation's first AI Coding Workshop in Brussels, led by Jonas Helming. The core message is moving from 'vibe coding' (prompt and hope) to 'Systematic AI Coding' — understanding how coding agents actually work and building repeatable workflows around them. Key insights include: LLMs are stateless and context is everything; efficiency degrades around 40% of context window capacity (the 'dumb zone'); time allocation shifts dramatically with AI (more design/review/debugging, less typing); and 'done' from an agent doesn't mean production-ready. Practical workflow advice covers when to start fresh sessions, how to compact context manually, and the importance of upfront design before prompting. The Eclipse Foundation's Open VSX and Theia IDE are also briefly highlighted.

Legibility of Effort · eieio.games

LLMs have eroded 'legibility of effort' — the ability to tell at a glance whether something took a human meaningful work. Heuristics like 'someone wrote this' or 'someone built this website' no longer reliably signal care or skill. Open source maintainers are struggling with AI-generated PR spam, and vibecoded websites share recognizable default aesthetics (warm-cream backgrounds, Claude-style layouts) that signal low intentionality. The author argues that making good things still requires hard decisions, but conveying that effort to others is increasingly difficult. In response, some people are gravitating toward hardware, physical hobbies, and intentionally rough aesthetics — paralleling how vinyl records re-added friction to music consumption. No solution is offered, but the cultural shift is framed as a hard, new, interesting problem worth watching.

I built the same app with Claude Design and its open-source rival, and the winner wasn't close

A hands-on comparison of Claude Design and Open Design, an open-source rival, using the same prompt and the same model (Anthropic's Opus) to build a local LLM benchmark tracker app. Claude Design produced a near-perfect first generation with polished inline editing, real-time theme switching, and a stable side panel. Open Design matched it on raw design output but struggled with a laggy, unreliable manual editing experience. The conclusion: both tools generate solid designs when powered by a strong model, but Claude Design has matured into a genuine editor while Open Design's editing UX still needs work.

Your developers are the attack surface now and vibe coding as a vulnerability | Tanya Janca

Tanya Janca, OWASP Top 10 Project Leader, explains why vibe coding has become a critical security risk and why developers themselves are now the primary attack surface. The newsletter also covers Meta's controversial restructuring (mandatory keystroke tracking, reassigning engineers to AI data labeling), Charity Majors' argument that AI-generated code demands more engineering discipline, the rise of domain-specific open-weight models like GLM-5.2, Microsoft potentially self-hosting DeepSeek for Copilot, and Anthropic's analysis of 400K Claude Code sessions showing humans drive 70%+ of planning while models handle 80%+ of execution.

Three Coding Interview Formats: And What Each One Actually Tests

A developer reflects on three coding interview formats encountered during a recent job search: LeetCode screens, take-home tasks, and vibe coding sessions. LeetCode is characterized as a preparation-heavy secret handshake that rewards pattern memorization over real engineering skill. Take-homes are seen as format-dependent, with quality varying based on how companies run them and follow-up discussions. Vibe coding — where candidates use AI tools while thinking aloud — is identified as an emerging format with real signal potential but significant variance across companies. Examples from Shopify, Meta, Stripe, Amazon, and Google show how differently companies are implementing AI-enabled interviews. The post argues that without explicit evaluation criteria, vibe coding risks becoming bias-driven, and that the ideal format is a realistic, collaborative, time-boxed problem followed by a discussion of decisions.

Ribbie turns real-time baseball stats into arcade-like, pixel art broadcasts

Ribbie is a fan-made website that transforms real-time MLB game data into animated 8-bit pixel art broadcasts, built by Eric Brownrout using Claude Code and Codex. What would have taken months was completed in a few weekends with AI coding assistance. The site pulls from MLB's public StatsAPI — the same one powering fantasy baseball platforms — to display live scores, pitching, batting, and base runners in a retro arcade aesthetic, including unique pixel-art representations of every stadium and player. Brownrout recently added fantasy baseball roster tracking and plans to add sound effects and fuller animations.

Vibecoding Malware Fuels a New Wave Of Stealth Attacks

Cybersecurity researchers at University College Cork have found that AI-assisted malware creation, dubbed 'vibeware,' poses a growing threat by generating functionally identical but structurally varied malicious code that evades traditional static detection methods like hash matching and YARA rules. The UCC team demonstrated that malware targeting Linux systems could be crafted with as few as two prompts using tools like Cursor, which did not restrict the requests. Experts from McAfee, Bitdefender, and ProjectDiscovery warn that LLMs lower the barrier to entry for threat actors, enabling even non-coders to build malware. The consensus is that defenders must shift to behavioral and AI-driven detection strategies. Google's discovery of PROMPTFLUX — malware that calls the Gemini API at runtime to rewrite its own code — illustrates how adaptive AI-generated malware could become.