There's a quiet revolution happening in our development team. It's not about the latest framework or tech stack; it's about how we work. As PK's lead mobile developer, I've been exploring collaborative AI coding tools like Cline, Cursor, and Windsurf alongside Monty (our Web & Platform Lead), who's been running parallel experiments in his team. Together, we're witnessing some fascinating shifts in how our developers create, collaborate, and feel about their craft.
— Chen Chen, Mobile Lead, PaperKite
Beyond vibe coding
While "vibe coding" – the practice of rapid prototyping and idea exploration with AI tools – has gained traction in development circles, we've found that the value of collaborative AI coding tools extends far beyond just quick iterations. At PaperKite, we're taking the speed and creativity of vibe coding and embedding it within a more holistic, human centred, developer-first approach. For us, that means using AI not just to accelerate ideas, but to foster deep learning, reduce friction in everyday tasks, and support developer wellbeing. It’s not just about building faster, it’s about building better, with tools that work for people, not the other way around.
The human in the loop
Let's get one thing straight: we're not replacing developers with AI. That's not what this is about. We're interested in something much more exciting: the amplified capabilities that emerge when talented humans pair with intelligent tools. What we're exploring is a true partnership: combining human creativity and critical thinking with an AI's vast knowledge and pattern recognition. Together, we're finding we can solve problems faster and more elegantly than either could alone.
Cline and similar AI coding assistants don't write perfect code on their own, and to be honest, neither do we. What happens instead is a conversation. Our developers prompt, refine, challenge, and direct. AI suggests, generates, and explains. Together, they create something better than either could alone.
What we've learned so far
Six months into our experiments with AI programming assistants, here's what we're seeing:
Speed without sacrificing quality
The productivity gains are real. Tasks that used to consume days - especially the tedious ones like writing tests, building documentation, or refactoring legacy code - can now be completed in hours. But speed isn't worth much if quality suffers.
The surprising discovery? Quality is improving. When our devs can focus more on architecture and problem-solving rather than typing out boilerplate, they make better decisions. The code is cleaner, more robust, and often more thoughtful.
As one of our senior mobile developers put it: "I'm spending less time writing code and more time thinking about whether it's the right code."
We've found these AI coding tools particularly shine when they're employed as an accelerant for tasks we already know how to do well, not as a replacement for thinking. I noticed Ross Jourdain from Abletech published some similar observations about AI coding tools that resonates with our experience.
Where Cline really shines
While we've experimented with several AI companions, Cline AI coding has stood out in a few key areas:
Handling complex app architectures: Cline seems particularly adept at understanding the relationships between different components in our mobile apps. In one project, we were refactoring a complex state management system, and Cline was able to trace through the data flow and suggest optimisations that other tools missed.
API integration: When facing poorly documented APIs (we all know the struggle!), Cline really shines. First, it generates modern documentation in minutes. Then, armed with this newly created context, it goes on to help efficiently complete the frontend integration. This two-step approach saved us considerable time during a recent payment gateway integration.
Context retention: Cline's persistent memory bank ensures excellent context retention. It remembers project context across sessions, so you can leave and return without repeating yourself. This is a major advantage over tools that lose context quickly and makes iterative refinement much more efficient.
Learning in both directions
In the mobile team, I've watched junior developers find a form of mentor in these AI programming assistants. They’ve been able to ask questions without fear of judgment, see multiple approaches to a problem, and understand why certain patterns are preferred over others.
Meanwhile, our more experienced devs are discovering new approaches and staying current with evolving best practices. There's something both humbling and refreshing about getting feedback from a tool that draws on diverse perspectives and coding wisdom from countless languages, paradigms, and problem-solving approaches (often suggesting solutions we wouldn't have considered).
Monty's been running parallel experiments with the web and platform team. We’ve been comparing notes and seeing where our experiences align and differ.
From Frustration to Focus: The Wellbeing Factor
Here's what really gets me excited: our developers report feeling less frustrated by the mundane aspects of coding and more engaged with the creative challenges - a key benefit of our developer wellbeing focus.
The mental load of context-switching, jumping between documentation and code editor, is significantly reduced. That mental bandwidth is freed up for deeper thinking and creativity.
We've noticed fewer frustrating afternoons fixing bugs and more enjoyable and balanced workdays. When your AI assistant can help you spot potential issues before you even run the code, you save yourself future headaches.
In a recent team retrospective, several developers mentioned feeling less stressed since we introduced these tools. That alone makes this exploration worthwhile to me.
Challenges worth solving
It's not all sunshine and automatically generated algorithms. We've encountered some real challenges in collaborative AI development:
Finding the right rhythm
Each developer on my team has had to find their own workflow with these tools. Some engage in constant back-and-forth dialogue, while others use the AI sparingly for specific tasks. There's no one-size-fits-all approach, and that's okay.
Monty and I have been deliberate about not mandating specific usage patterns. Instead, we've created space for experimentation and knowledge sharing, letting each developer discover what works for them. We're looking forward to sharing more about our different approaches in the future.
Maintaining coding muscles
There's a legitimate concern that over-reliance could lead to skill atrophy (with increasing discussion about generative AI's impact on critical thinking). I definitely think that there is a trap in letting AI write too much of your code for you. I'm mindful of this and encourage my team to understand the code they're implementing, not just accept suggestions without review.
Our code reviews have always emphasised optimisation and learning, not just functionality. With AI-generated code, we've simply expanded this focus, encouraging deeper discussions about why certain approaches work and how they integrate with our broader architecture. If anything, these tools have enhanced our learning culture by giving us more to discuss and explore together.
Trust and verification
Every suggestion requires verification. While the tools are impressive, they can still produce elegant-looking code that's subtly wrong. My developers have developed a healthy skepticism that balances AI assistance with human expertise.
One practice I've implemented is having AI-generated code reviewed with extra scrutiny. Not because it's inherently suspect, but because understanding externally generated code is a different skill than understanding code you've written yourself.
Transparency and security
The more context we give these tools about our codebase, the better they perform. But this can’t come at the cost of your security or intellectual property.
Cline offers two compelling advantages here. First, it doesn't transmit your code to external parties, keeping your intellectual property secure. Second, as an open-source tool, its code is fully transparent, allowing anyone to verify these security claims rather than taking them on faith. This combination of privacy and verifiability gives us confidence that our proprietary code stays exactly where it belongs.
When paired with a trusted, enterprise-level AI model, the system provides robust security while still delivering powerful assistance. Cline also incorporates an ignore configuration file (similar to Git), allowing us to explicitly define boundaries—telling the AI assistant which files or folders are off-limits. This gives us fine-grained control over what information we share, protecting sensitive code with confidence.
Evolving our approach to pair programming
One fascinating development has been how these tools have influenced our pair programming practices. Pair programming feels more effective when there are three of you, and one of them is an AI. I've seen this firsthand in our collaborative AI development process.
When two developers collaborate with Cline or similar tools, their conversations become richer. The AI acts as a third perspective. AI is sometimes a mediator, sometimes a challenger, and sometimes simply a rapid prototyper that allows the human developers to focus on higher-level design decisions and tradeoffs.
In one recent mobile sprint, we tackled a particularly thorny state management problem by having two developers pair with Cline. The AI helped them rapidly prototype three different approaches, allowing them to compare trade-offs in a way that would have taken days without assistance.
What's better for people is better for code
At PaperKite, we've always believed that uplifting people leads to better outcomes. Our experiments reinforce that philosophy. When we give our developers tools that reduce friction and frustration while amplifying their capabilities, everyone wins; our team, our clients, and the people who use the digital products we create.
We're still very much on this journey, but I'm excited about where it's heading. The future of development isn't about replacing humans with AI, it's about creating powerful partnerships that bring out the best in both.
These tools are changing how we work, but our core values remain: creating digital experiences that uplift people and transform their everyday. Now we're just doing it with some powerful (and elegantly designed) AI assistants by our side.
Chen is the Lead Mobile Developer at PaperKite. He's been exploring how AI-assisted coding tools can enhance developer wellbeing and productivity while maintaining PK's commitment to quality and craft. Monty, our web development lead, will be sharing his experiences in an upcoming post.
Want to chat about how we're evolving our development practices? We'd love to kōrero about it. Get in touch and let's explore how these approaches might benefit your digital projects.
Take your first steps into AI with confidence.
Let's chat about what's possible.