Transition to an AI-supported product discovery and delivery approach

Organisations face a crucial challenge: how to harness the power of artificial intelligence effectively while maintaining the human creativity and expertise that drives true innovation. Our team’s journey with AI adoption offers valuable insights into this balancing act; one that has transformed our productivity, quality, and impact, even through tough times with resource constraints.

In this case study, I’m sharing how our lean product team of just 6 engineers has built a successful platform serving millions of users. I’ll explore our transition to an AI-supported Product Discovery and Delivery approach, the methodology behind our implementation decisions, and the remarkable results we’ve achieved through this journey of strategic integration of artificial intelligence into our workflow.

The catalyst: Necessity drives innovation

Like many teams navigating today’s economic realities, we faced a significant reduction in our workforce—30% fewer team members with the same (or greater) expectations for delivery. As engineers passionate about problem-solving and technology experimentation, we saw this challenge as an opportunity to reimagine our approach to work. 

When AI tools began emerging on the market, we didn’t rush to integrate them indiscriminately. Instead, we took a methodical approach guided by two fundamental principles:

1. Technology immersion: Actively experimenting with new technologies to understand their capabilities and limitations firsthand.
2. Problem clarity: Gaining deep insight into our specific challenges and identifying inefficiency points that technology could address.

As a Physics Engineer, I’ve always approached challenges using the scientific method—observe, hypothesize, test, and iterate. This mindset proved invaluable as we tackled AI integration. We actively experimented with new technologies to understand their capabilities and limitations firsthand, and worked to gain deep insight into our specific challenges, identifying inefficiency points that technology could address.

Finding the right AI applications: A strategic approach

Working efficiently with AI isn’t about blindly integrating it into every process—it’s about understanding where it can genuinely enhance human work. I like to think of it as identifying the right problems to solve that will bring the most value to both our team and our users.

We started by identifying tasks that fit specific criteria:

• Repetitive or tedious work that drained creative energy
• Tasks requiring processing of large volumes of data
• Documentation generation and note-taking during meetings
• Code assistance for routine implementation

This targeted approach immediately yielded results that far exceeded expectations. Despite a 30% reduction in team size, our Product Team not only maintained performance but also significantly over-delivered across multiple fronts. Most impressively, we achieved the majority of our annual Product Team OKRs, including:

• Improving platform availability and reliability
• Enhancing user retention metrics
• Significantly increasing app store ratings
• Enabling substantial growth in our user base

Much of my time as a Product Lead is dedicated to observing the world around me, crafting and refining our Product Strategy, constantly thinking ahead to create and capture value for users and the company. With AI support, this process became even more effective, allowing us to focus on the high-level strategic work while delegating more routine tasks.

These improvements didn’t go unnoticed. Leadership recognised not just quantitative metrics but qualitative enhancements in implementation speed, coordination quality, and strategic alignment. Ultimately, these achievements contributed directly to our successful Series B funding round—investors recognised the operational excellence that our AI-enhanced approach enabled.

The success has been so pronounced that we’re now expanding this methodology to other departments across the company, applying these product-led efficiency principles. By scaling what worked for our team, we’re creating a multiplier effect that promises to further accelerate our growth trajectory.

Engineering excellence: Team-specific AI implementation

Backend Development: The Evolution of Our Approach

Our backend team’s AI journey illustrates the importance of finding the right tool for the right problem. Their initial attempts with various AI chat interfaces for code assistance revealed significant limitations:

• Direct chat interfaces struggled with complex code contexts, and we struggled to give sufficient context by simply explaining
• Providing sufficient context was time-consuming
• Solutions often miss the mark without codebase understanding

The breakthrough came with adopting Cursor—an AI-enhanced development environment that integrates with our existing workflow while providing crucial context awareness. While not perfect, it now handles approximately 70% of routine coding tasks, allowing engineers to focus on fine-tuning solutions based on customer needs and product feedback.

The result? A 60% performance improvement in feature delivery time to achieve specific outcomes while keeping quality standards high / same quality as before. This combines a more effective product discovery where human engineers have a critical role in idealising solutions that match the problems we are trying to solve and users’ needs, as well as using AI to accelerate delivery with high-quality code standards from our top-notch engineers.

The engineer’s role has evolved to focus on problem identification and creative solution development—using AI as a powerful assistant rather than a replacement. From my experience, this shift in focus is critical—it’s not about replacing humans but enhancing their capabilities to deliver more value.

Infrastructure team: From frustration to focused productivity

Our infrastructure team’s journey with AI began with typical growing pains. Early experimentation felt frustrating as the team grappled with fundamental questions: which use cases were appropriate for AI assistance, and which were better left to traditional methods? As one team member noted, “In the beginning, it’s frustrating when you don’t know the models, and we were starting without knowing how or for what to use AI because it’s not for everything”. For example, as we speak we’ve been trying to generate tests for a Supabase Edge Function (using Deno), and none of the current models could give us a functional implementation. Another example was we’ve had some AI models circling trying different implementations of data streaming in a Next.js app - and failing -, just for humans to realize that the functionality wasn’t necessary in the first place, the requirements were just wrong.
These two examples give us two very powerful lessons: 

1. Understanding limitations is a critical step for technology adoption. 
2. AI cannot solve problems when the problem itself is not well defined.

In these two cases, talented human engineers need to be able to understand limitations and define the problems. This is where we need to spend our time. We cannot blindly trust in AI, it can interpret the problem wrongly and suggest solutions that are not ideal! Independently of the wrong or right requirements, there are tasks that even with different models, AI cannot solve, and will just circle around until you understand you need to be the one solving those. Knowing all of this is priceless for our engineers. 

Their approach evolved through several key phases.

1. Model differentiation: The team discovered that different AI models excel at different tasks. Some provide exceptional reasoning capabilities, while others excel at explaining concepts or offering alternatives. Understanding these distinctions allowed them to select appropriate tools for specific infrastructure challenges.
2. Prompt engineering: Developing effective prompting techniques represented a significant skill advancement. The team learned to frame requests that produced useful outputs while maintaining control over critical decisions.
3. Task identification: Rather than defaulting to AI for everything, the infrastructure team developed a nuanced understanding of which tasks benefit most from automation. They found particular value in delegating repetitive tasks while maintaining human oversight of architecture and design.

Today, the infrastructure team uses AI as a productivity partner primarily for infrastructure-as-code work. They create the architectural skeleton and allow AI to handle tedious implementation details in tools like Ansible and Terraform. This approach preserves their engineering craft while eliminating low-value repetition—a balance that aligns with their technical standards and team values.

The results speak for themselves: with just a single engineer managing the entire infrastructure that serves millions of users, we’ve not only enabled growth as planned but maintained exceptional reliability with 99.99% platform availability. This level of operational excellence with such lean staffing would have been unimaginable without our strategic AI implementation.

Mobile development: Balancing innovation with responsibility

Our mobile team takes a carefully calibrated approach to AI adoption. Their experience highlights important considerations:

• Human supervision remains essential—engineers must understand and take responsibility for all code
• Critical thinking skills must be actively maintained
• Tool selection must align with specific development environments

Initially, they faced significant challenges. Xcode (Apple’s official IDE) and Android Studio lag considerably behind in AI capabilities compared to newer AI-enhanced code editors. The immediate feedback loop that makes Xcode valuable for iOS development wasn’t fully replicated in early AI-enhanced environments, and AI suggestions weren’t always valid in mobile contexts.

Rather than seeing these as permanent blockers, the team viewed them as research opportunities and began innovating their workflow. Their breakthrough came with a sophisticated integration approach.

“The mobile team started to use Cursor AI, which provides access to all of the great features like smart code completion, smart text predictions, chat, code agents, etc. to build, run, debug and test mobile apps,” explains our mobile lead.

The key was recognising that mobile developers require strict environments to ship apps to stores—particularly for iOS, which heavily relies on Xcode. Their solution was transforming Cursor (based on Visual Studio Code) into a robust mobile development environment through strategic extensions.

“By using Sweetpad extension, I can have Swift language code completion with iOS SDKs, libraries and components. I can build and run iOS apps directly from Cursor, control iOS simulators from the editor, and even debug seamlessly with CoreLLDB integration” notes a senior iOS developer. “This reduces the previous weaknesses of replicating Xcode mechanics for iOS development. And the best part is that I can use all the AI features Cursor provides to excel in mobile app development.”

This approach has allowed the mobile team to overcome their initial limitations while preserving the strict requirements for mobile development. The impact has been remarkable: not only are they shipping code much faster, but the quality and effectiveness of their work have improved dramatically. This is reflected in key proxy metrics, particularly user retention, and has helped us achieve an impressive 4.9 rating on the App Store—allowing us to go toe-to-toe with our most significant competitors in the market.

I believe this success shows how important it is to adapt tools to your specific context. No off-the-shelf solution will perfectly match your needs—the value comes from your team’s ability to customize and integrate technologies to solve your unique challenges.

Beyond code: Documentation and planning

Other of our most efficient AI applications have been in traditionally time-consuming administrative tasks:

• Sprint planning documentation
• Feature specification development
• Meeting notes and summaries
• Discovery documentation

In each case, we maintain a critical human element—an assigned owner who reviews, edits and enhances AI-generated content. Our Product Managers, in particular, ensure that AI-assisted documentation maintains the “sparkle of creativity and customer empathy” that differentiates truly valuable products from merely functional ones. It also allows us to better manage our teams and understand how they are feeling and how we can help them. 

One of the most significant achievements in this area was the ability to consistently analyse insights from tens of thousands of customer reviews and feedback during product discovery phases. This scale of customer insight analysis would have been prohibitively time-consuming using traditional methods. Still, AI tools enabled the team to identify patterns, prioritise feedback, and make data-driven decisions while maintaining a deep understanding of user needs.

The true value proposition

Our experience has reinforced a critical insight: businesses rarely fail because coding isn’t fast enough—they fail because value creation isn’t aligned with company goals and customer needs. AI helps us achieve this balance more effectively by handling routine tasks and amplifying our human capabilities.

The result is a team that can focus more intently on outcomes rather than outputs, creating solutions that genuinely address customer needs while advancing strategic objectives. This approach has resonated with investors as well—we recently closed our Series B funding round, a significant validation of both our product vision and our efficient, AI-enhanced operational approach.

How many of you have experienced working in a company where the majority of time is spent on routine tasks rather than strategic thinking? Too many, I suspect! The real power of AI is in freeing your team to focus on the most valuable aspects of product development—understanding users, crafting strategies, and designing solutions that truly matter.

Our ongoing journey

This transformation hasn’t been a one-time implementation but an evolving process of experimentation, learning, and refinement. We continue to evaluate new AI capabilities, always with a clear focus on how they might enhance—rather than replace—the human expertise that remains our most valuable asset.

We are now exploring adding AI standards to each of our projects, similar to any other documentation that teams can use to help their work. This includes encouraging everyone to share their setups inside a project, documenting the prompts that bring better results for different tasks, and making sure that team members are consistently sharing new tools they are exploring and the results they get. Our weeklies and retros now have a space for sharing, a space for teams to discuss their different approaches and raise the bar on our standards. This knowledge sharing creates a multiplier effect, allowing us to build on each other’s discoveries rather than having each team member reinvent approaches in isolation.

A critical element of our success has been creating space for experimentation and even failure. Your teams need to fail, experiment, and have time and incentives to do that—that’s the only way we go forward and get the passion and will to cooperate with AI rather than blindly use it or not using it at all due to fear. By encouraging this exploratory mindset, we’ve built team members who approach AI as collaborative partners rather than either magic solutions or existential threats.

By maintaining this balanced approach, we’re not just becoming more efficient—we’re positioning ourselves as the best AI-enabled team in our field, creating exceptional value through the thoughtful fusion of human creativity and artificial intelligence.

The key to our success has been understanding that AI is a powerful tool, not a silver bullet. By carefully selecting where and how to apply it, maintaining rigorous human oversight, and continuously evaluating its effectiveness, we’ve created a model for AI adoption that enhances rather than diminishes the human elements that make our work truly valuable.

I don’t intend to give you a step-by-step recipe for AI implementation—the process is too complex and different from company to company. Instead, I want to give you a framework and values to work with. The beauty of this approach lies in its adaptability to your specific context, the problems you’re trying to solve, and the unique skills of your team.