2025 June Digest

Recently, I’ve been experimenting with AI coding assistants like Cursor to build some side projects (more on that soon). This naturally sparked my curiosity about how others are using these tools and whether any best practices are starting to emerge. My June digest highlights a few noteworthy articles and YouTube videos I came across during this exploration.

Blog Posts

Coding Agent Has Crossed the Chasm

In this article, the author describes a shift in perspective - from seeing AI coding tools as helpful novelties to viewing them as essential parts of the development workflow. Tools like Claude Code, Codex, and Cursor have evolved beyond simple autocomplete to autonomously completing tasks, debugging, and reviewing code - acting like capable interns.

This transformation has significantly boosted productivity, allowing the developer to delegate routine tasks and focus more on architecture, design, and problem-solving.

A key insight is that the effectiveness of these tools depends heavily on providing the right context—prompting is now a crucial skill. While AI can amplify both strengths and weaknesses, it’s most powerful in the hands of experienced developers who can guide and review its output.

Despite concerns about trustworthiness and job displacement, the author argues that AI is augmenting—not replacing—developers, and we’ve entered a new era where human-AI collaboration is becoming the norm.

How to Vibe Code as a Senior Engineer

Vibe coding is a new way of building software where you use one of the latest AI models to do most of the work. In this article, the author offers a few tips and tricks on how to vibe code effectively. The short takeaway is that context is everything when using LLM for coding.

• Use a strong scaffold: Start with a monorepo scaffold that includes examples to help the AI generalize patterns effectively. This is especially useful for bootstrapping a new project.
• Define strong rules: Use .cursor/rules in Cursor to enforce consistent project conventions (e.g., how to test, typecheck, avoid local server calls).
• Document project structure: Include clear explanations of directory layouts, libraries used, and coding standards.
• Open all relevant files: Especially type files and then use “Add All Open Files to Context” in Cursor.
• Don’t skimp on context: It’s worth the token cost—AI needs full visibility to be helpful.
• Prefer monorepos: Easier for AI to access full codebase context and avoids cross-package build/lint issues.
• Favor short iteration loops: Write, lint, test, repeat—don’t let the AI generate too much at once.
• Prompting tips: start with a detailed plan, be specific about output, give examples, use constraints, keep scope narrow.

YouTube Videos

Lenny's Podcast: Interview with Guillermo Rauch

This YouTube podcast episode features an insightful conversation with Guillermo Rauch, CEO and founder of Vercel and its AI-powered website builder, v0. The discussion centers on how v0 is democratizing product development, enabling more individuals to build and ship ideas with unprecedented ease, even those without traditional coding skills.

Rauch emphasizes that AI is transforming software creation, moving programming tasks towards "translation" while highlighting the enduring value of understanding foundational systems and cultivating "taste" through exposure and iteration. Some highlights:

• The Diminishment of Translation Tasks: Many traditional programming jobs involve “translation tasks,” such as converting design specs into code. AI excels at this—rapidly transforming high-level intent or designs into precise, executable code. As a result, the need for human developers who primarily perform this type of conversion is diminishing.
• Future of Engineering: Engineering is shifting from rote memorization to conceptual understanding. Developers will rely less on remembering syntax and more on grasping how systems work holistically—from databases to frontend architecture. This shift enables engineers to focus on more complex, creative, and high-impact problems.

Important Skills For the Future of Software Engineering

• Conceptual clarity: in logic, math, and systems thinking.
• Eloquence in prompting: the better you can express intent, the better the AI performs. This includes understanding how language tokens affect outputs
• Taste and intuition: cultivated by experimenting with products and observing real user behavior. This will improve with exposures overtime.
• Problem-solving resilience: the ability to debug, adapt, and get unstuck will remain as crucial as ever.

Cursor Tutorials for Beginners

This YouTube video covers key strategies and insights for effectively utilizing Cursor AI, drawn from an expert discussion on maximizing its potential for code generation and development. The core philosophy revolves around providing AI with maximum context and treating it as a co-pilot rather than an autonomous boss.

Here are some of the highlights:

Planning First

• Effective coding starts with thorough planning.
• Adopt a developer mindset that emphasizes thinking through what you’re going to build and how it should look before writing any code.
• Use tools like visual planners, v0.dev, or simple sketches to iterate on design ideas early in the process.

Provide Context to Cursor

• Use .cursor_rules: Visit cursor.directory to find pre-written prompt templates. These can be copied into a .cursor_rules file in your project’s root to guide the AI’s behavior.
• Customize for Your Stack: If your stack isn’t listed, copy a template and ask a model like Claude or GPT to adapt it for your specific technologies.
• Tag Documentation: Use Cursor’s docs tagging feature to provide URLs to official tech docs (e.g., nextjs.org, supabase.com). This improves the AI’s accuracy and relevance.

How to leverage AI

• Understanding Code: Ask Cursor to explain code “like I’m a beginner” to clarify flow, logic, and structure—especially helpful in large or unfamiliar codebases.
• Learning Concepts: Use AI to teach programming concepts on demand. When confused by a term or snippet, ask “what does this mean?” and iterate from there.
• Generate Documentation: Let AI handle documentation for existing code—an often tedious but necessary task that it excels at.