A few months ago, I shared a video showcasing how I leverage AI to supercharge my writing of apps. In the fast-paced world of AI development, “a few months ago” feels like an eternity! This time, my study buddy and I sat down for a Sunday deep dive to unpack just how much has transformed since then, especially with tools like Cursor. The biggest shift we’re seeing in IDEs is the profound impact on developer flow, particularly through how we inject context and the rise of truly agentic workflows. This an informal chat that we decided to record and I’ve provided links to relevant time slots for the areas you’re interested in.
Gone are the days of purely manual input. Now, you can calibrate your AI assistant from highly agentic — where the model takes the initiative, intelligently building context by searching your codebase, calling MCPs (Model Context Protocols), and utilizing various tools to achieve your goal — all the way to manual, giving you granular control over every piece of context passed to the model. This evolution is powered by continuous improvements in models, with certain ones proving exceptionally better for specific tasks. We also explored how Cursor rules let you define custom behaviors, like adhering to specific coding styles or using Cursor’s docs feature to search documentation online either through llm.txt or a full WebCrawler that locally indexed, providing up-to-date information that might not even be in the model’s training weights! And for many developers, the Command Line Interface (CLI), with powerful tools like Claude Code, is really starting to take the lead in this new era of AI-driven development.
Unfortunately, we didn’t get a chance to dive into how to effectively use an LLM for coding. One of my favorite approaches is applying George Pólya’s four-step problem-solving framework (Understand, Plan, Act, Review) as structured prompts to guide AI-assisted development. This systematic approach to breaking down coding challenges might be perfect for a future video. Some of these techniques can really change the experience of working with AI, it is a new skill to learn. Not just for coding, but for any task where you want to leverage AI to help you.
Beyond Agentic: Cursor’s Flexible Modes
What truly impressed us about Cursor is its versatility in adapting to different developer preferences. It’s not a one-size-fits-all solution; instead, it offers distinct modes to match your comfort level and the task at hand:
- Agent Mode (4:05, 5:34): This is where Cursor truly shines as an “agent,” taking your prompts and autonomously executing tasks, including making tool calls to get the job done.
- Ask Mode (23:29, 24:07): For those who prefer a more deliberate pace, Ask Mode presents the AI-generated code in a writing box, allowing you to review and refine it before it automatically creates or modifies files. It’s a fantastic way to slow down and truly understand the AI’s suggestions.
- Manual Mode (24:45): If you want complete control, Manual Mode ensures the AI only uses the context you explicitly provide, relying purely on its model weights and your typed input.
- Background Mode (20:02, 25:15, 53:42): This is a game-changer for larger projects or when you need to run multiple tasks. Cursor can handle complex operations in the background, potentially on a remote VM, freeing you up to continue other work.
Mastering Context: Guiding Your AI
The power of AI lies in its understanding of context, and Cursor provides robust ways to ensure your AI assistant is always on the same page as you. You can precisely guide the AI by adding specific lines of code, entire files, or even whole folders to the chat context (32:12). Beyond your codebase, Cursor’s Docs Integration is a standout feature. You can simply paste a URL, and Cursor will index the entire site, allowing the AI to reference up-to-date documentation (35:10, 35:34, 39:18). This is invaluable for ensuring the AI works with the latest library versions or framework best practices.
Seamless Editing and Debugging
Cursor isn’t just about generating code; it’s about making the entire development lifecycle smoother. Before accepting any AI-generated changes, you get to review detailed diffs, make manual edits, or even reject the changes entirely (12:40, 12:58). This level of control is a significant improvement over earlier AI coding assistants. For quick tweaks, Inline Editing (Ctrl+K) allows you to modify specific lines of code directly without leaving your current view (64:25). And if you’re looking for advanced auto-completion, the Cursor Tab feature is described as “GitHub Copilot on steroids,” intelligently suggesting code as you type (63:29, 63:45).
Debugging also gets a major upgrade with Cursor’s Terminal Integration. The editor can read error logs and other crucial information directly from the integrated terminal, providing the AI with the necessary context to help you troubleshoot and fix issues more efficiently (48:38, 55:17). Plus, the Checkpoints feature (16:11) offers a comforting safety net, allowing you to easily roll back to previous versions of your code.
The Evolving Landscape of AI in Development
Our chat wasn’t just about Cursor; it naturally expanded into a broader discussion about the future of AI in coding. We touched upon how tools like VS Code with GitHub Copilot (2:28, 2:39) and Codeium (or CodeX as it was referred to) (19:09, 20:31) compare. We also explored the practical considerations of subscription costs for various AI tools, including Cursor, OpenAI, and Google Gemini (16:26, 16:53).
Perhaps the most profound insight was the evolving role of the developer. While AI can automate vast amounts of coding, our role shifts from simply writing code to understanding the problem deeply, effectively guiding the AI, and rigorously ensuring the generated code meets all requirements (15:14, 68:10). This emphasizes the growing importance of “context engineering” – the next evolution of “prompt engineering” (14:13, 57:32, 71:36). We even had a brief, fascinating detour into OpenAI’s voice chat feature, showcasing its potential beyond coding (27:04, 28:28, 30:25).
Our Takeaway
The video truly highlights how AI tools like Cursor are not just incremental improvements; they are fundamentally changing developer workflows. It’s an exciting time to be a developer, with powerful agents at our fingertips. The key, as we discovered, lies in understanding how to effectively harness these tools by providing the right context and adapting our own processes to collaborate seamlessly with AI.