Best of AI-Assisted Development — February 2026

1. 1
   Article

   Cloudflare·13w

   How we rebuilt Next.js with AI in one week

   Cloudflare engineers rebuilt Next.js from scratch on top of Vite in under a week using AI (Claude via OpenCode), producing an open-source drop-in replacement called vinext. It builds production apps up to 4.4x faster with Vite 8/Rolldown and produces client bundles 57% smaller than Next.js 16. vinext deploys directly to Cloudflare Workers with a single command, supports both App Router and Pages Router, includes ISR via KV, and covers 94% of the Next.js 16 API surface backed by 1,700+ Vitest tests and 380 Playwright E2E tests. A novel 'Traffic-aware Pre-Rendering' (TPR) feature uses Cloudflare zone analytics to pre-render only the pages that actually receive traffic, avoiding the linear build-time scaling problem of large Next.js sites. The project cost roughly $1,100 in Claude API tokens and raises broader questions about which software abstractions exist for human cognitive limits versus genuine architectural necessity.

   131

   1
2. 2
   Article

   Kitty says hi.·16w

   Cursor for UI work

   Cursor AI can rapidly bootstrap React applications and handle initial setup tasks much faster than manual coding, but struggles with code quality, component architecture, and creative UI design. The tool tends to create monolithic components with poor logic encapsulation, duplicated code, and limited reusability. While it excels at generating B2B dashboard-style interfaces, it lacks creativity for custom designs and requires significant manual refinement. Best results come from giving narrow, precise tasks rather than broad directives, particularly around component reusability and logic separation.

   96

   4
3. 3
   Video

   AICodeKing·16w

   Codex Desktop App + Free GPT-5.2 Codex (Tested): Is OpenAI now copying Conductor,Commander?

   OpenAI launched a desktop app for Codex (GPT-5.2), available free for a month and currently macOS-only. The app provides a graphical interface with features like skills, automations, and work trees, drawing comparisons to Conductor. However, the review highlights numerous UI/UX issues including inconsistent design, buggy interfaces, poor context handling, unintuitive controls (like the plan mode toggle), and problematic VS Code integration that spawns multiple instances. The reviewer suggests competitors like Verdant offer superior agentic interfaces despite OpenAI's resources.

   43

   1
4. 4
   Article

   Apple Developer·16w

   Hello Developer: February 2026

   Apple announces agentic coding capabilities coming to Xcode, alongside upcoming developer events including a code-along session on coding intelligence and the Swift Student Challenge. The update also highlights new learning resources covering SwiftUI, design, and security topics for Apple platform developers.

   32
5. 5
   Article

   Collections·13w

   Ladybird Browser Transitions to Rust with AI Assistance

   The Ladybird browser project is migrating its codebase from C++ to Rust, beginning with the JavaScript engine LibJS. Lead developer Andreas Kling used AI coding tools—Claude Code and Codex—to accelerate the translation, converting ~25,000 lines of C++ to Rust in two weeks, a process that would otherwise have taken months. The translated code produces byte-for-byte identical output and passed all regression tests. Rust was chosen over Swift due to better C++ interoperability and a more mature ecosystem. C++ and Rust will coexist going forward, with the migration proceeding gradually under human-directed architectural oversight.

   9

   3
6. 6
   Article

   The Miners·13w

   Agentic Engineering Is Just Good Engineering (With a Better Driver)

   Agentic Engineering is proposed as a deliberate middle ground between rejecting AI entirely and blindly vibecoding. The author argues that developers must adopt AI tools or risk being outpaced, but that using them well requires planning, incremental steps, guardrails (linters, tests, git hooks, context files like CLAUDE.md), and deep technical fundamentals. The agent is framed as the driver in a pair-programming model, with the developer as the navigator making all meaningful decisions. Deep knowledge of systems, architecture, and product context becomes more valuable, not less, in this paradigm.

   10

   1