Having explored the challenges of AI coding assistants in my previous article, I'm now turning to the substantial benefits these tools can bring to a developer's workflow. After months of intensive use, integrating Cursor.ai as my primary code editor and Claude AI for both direct and in-editor assistance, I've discovered several ways these technologies genuinely enhance productivity and creativity.
While it's crucial to approach AI tools with a clear understanding of their limitations, it's equally important to recognize their transformative potential when used strategically. This article highlights the positive impacts I've experienced across various aspects of development work, from initial project scoping to complex implementation tasks. My goal is to show how these tools, when used thoughtfully, can augment rather than replace human expertise.
Disclaimer
Using these AI tools can potentially expose sensitive information about your organization, not only at the code level but also regarding structure and personnel. In theory, your API keys and other information could be stored or trained on, along with potentially sensitive data about users, team members, and other exploitable information.
When using any AI tool, always operate under the assumption that the information you're sharing is not private or secure.
If you're a junior dev
Please first read the Challenges of using AI before diving into this article. I want to emphasize that solely relying on Cursor and AI to code will likely impair your ability to learn coding fundamentals and maintain production-ready codebases. Look to your team for guidance and initially use Cursor as a replacement for Stack Overflow and for templating your code.
Cursor.ai
Cursor has settings that allow you to control what information is sent externally for AI assistance. Please review these settings and ensure certain filters and files are marked to prevent them from being sent to the server. Also check the code retention settings in your Cursor account to adjust how long your code queries are stored.
Claude.ai
Because Claude.ai via the browser requires specific queries and manually providing the code you want analyzed or modified, you have more control over what information is shared. However, the code is still retained on Claude's servers for up to 30 days, so ensure sensitive structures and information are not being exposed.
Clarification
When I mention Cursor.ai below, I'm referring to the Claude.ai agent or chat functionality within Cursor, not GPT. When I mention Claude.ai, I'm referring to the browser version of the service.
Over-using AI tools
As we will explore in these articles, there are compelling reasons not to overly rely on AI tools for code creation. AI should be used strategically—as a tool to automate mundane tasks, gather insights, and serve as a sparring partner. Excessive dependence on AI to write and fix your code will ultimately create problems for both you and your organization.
Positives of Using AI
In the previous article, I did a deep dive into the challenges and drawbacks of using Cursor and Claude in my regular work. In this article, I'll explore the positive aspects and ways that Cursor and Claude are accelerating my development process and helping me adopt new projects and code more efficiently.
Scoping a project
The first major benefit that Claude.ai has provided is the ability to quickly generate an initial project scope document. It reminds me of important details that are sometimes overlooked when estimating the time and cost of a project or task.
Please write a specification document for a pizza delivery website, has user accounts, manage orders, real-time order tracking, menu management, payment handling in the Netherlands and SEO optimizations and powered by Prismic
The prompt will outline nearly everything I need to consider, including services, integrations, and easily overlooked details when focusing on complex features, such as SEO.
By using Claude as a foundation to template large tasks, I can provide more accurate and meaningful estimates faster and rely on fewer iterations during scope and planning meetings.
Notice how I avoid including specific details or features that might be sensitive to the project—those are elements I'll fill in manually.
Thinking of the problem more deeply
AI works well because it demonstrates the 'garbage in, garbage out' principle—if you don't provide clear and consistent information, Claude will likely return poor results.
This dynamic encourages you to think more deeply about the problem and how to approach solving it. The quality of the answers you receive directly reflects your approach to framing the problem.
If we first look at a bad prompt: I need to fix the product listing page, it first shows out of date content and then shows the real content afterwards. This explains the problem, but vaguely, and might not lead to the right solution.
Let's look at explaining the issue better: I want to refactor this screen as it relies on stale data initially while the background fetch is happening. Whenever there are query parameters in the URL, I want the client-side fetch to complete first before showing any content. Please use the loading component in our /core/components folder - /core/components/loader.js
Here, we've given the AI more context about the root of the problem, the considerations it needs to make, and how to handle the request gracefully.
We also avoid flooding the request with too much information by scoping the context to only the necessary files.
Considering improvements
This section should be taken with a grain of salt, as there's often missing context and personal preferences that significantly affect what constitutes an actual improvement.
That said, what I appreciate about Cursor AI is the ability to provide it with the global context of my codebase and ask for specific improvements: How can I improve the SEO of this project? The AI then returns useful pointers and suggestions. For these types of requests, I find the Chat/Ask option more effective than the Agent.
This is still a sensitive area—a suggested change might technically be an improvement but may not yield any meaningful benefits. People can easily argue, "Claude suggested it, so it must be the way forward." Consider a simple line like: const ids = projects.map(id => id); In React, you might want to memoize this to prevent recalculation on every render, but if your projects array only contains about 10 entries, the cost of memoization might exceed the cost of the operation itself.
I list this as a benefit because of the "considerations" aspect—in a healthy, balanced team, these suggestions can spark valuable discussions leading to knowledge sharing and identifying areas for improvement within your team's practices, such as standards to include in your definition of done.
Onboarding
The feeling of entering a new codebase has never really changed for me, even after 10+ years of software development. Each project contains a history of decisions, conversations, and legacy code all bundled together.
I strive to do meaningful work from day one, but onboarding can be time-consuming regardless of experience. I've found Cursor to be an excellent tool for making this process faster and more efficient.
Here are some commands that can accelerate onboarding:
Please provide for me in an MD file a breakdown of this project, folder structure, links to document files, component breakdowns and where they're used
I'm looking for how the data gets into the shoppingCart.js file from the header, can you explain the prop drilling path or how data is shared here?
Using these alongside your regular onboarding process allows you to ramp up quickly and gain a better feel for the codebase and project.
Learning
This section deserves a quick disclaimer: the benefit of using AI to learn coding largely depends on the mentality of the person using it.
Using Cursor or AI to simply generate a working project without taking time to understand what's happening, why it works, and whether the AI-suggested code is appropriate will leave you at a disadvantage.
If you're using Cursor to adopt a new library, framework, or language, approach it the same way you'd learn anything else: focus on understanding the fundamentals.
Use Cursor to help you understand why certain approaches don't work, why others do, and what alternative solutions might solve your problem. This creates a learning experience similar to having a semi-experienced developer guiding you along the way.
Personally, I had some experience with PHP 8+ years ago. In recent months, I've been picking it up again, and Cursor has significantly helped me understand the changes in PHP and modern approaches that have evolved over those 8 years.
Back-end & Design systems
One of the most valuable outcomes from incorporating AI into our team has been its impact on cross-functional collaboration. Cursor has demonstrated a remarkable ability to understand our design system components and accurately apply them to new screens and functionality. This has created a bridge between our front-end design principles and back-end implementation that wasn't previously possible.
With this capability, our back-end developers can now rapidly build out new features with minimal guidance from the front-end team. The result is a more efficient workflow where front-end developers can focus their expertise on iterative UX improvements and the ongoing maintenance and evolution of our design system. This division of labor has accelerated our development cycle while maintaining visual and functional consistency across the application.
In ending
AI is like an additional junior developer
I've had a positive experience using Cursor as if it were another junior developer on my team. I can explain problems or provide templates and allow it to make initial multi-file changes while I focus on other tasks like checking emails or speaking with colleagues.
It has enabled me to develop projects faster and with greater confidence, though not without the limitations and challenges I've covered here.
AI currently does not replace developers
At present, AI doesn't replace developers—it transforms how we develop, how we articulate and conceptualize problems, and how we approach building solutions.
