Going AI-Native: A Product Development Studio’s Journey
“The leaderboard in your industry is going to shift violently over the next few years.” – Ravi Gupta, AI or Die
We’ve seen it first hand at Codewalla. As a product development studio, we’re always experimenting with bleeding-edge tech—anything that might help us deliver on our core mission: help new products launch faster. So when GenAI arrived, we dove in. We built prototypes. Ran hackathons. Shipped early POCs. Put features into production. Iterated on product increments making the features more useful, deeper, enterprise grade, running in our VPC, in other words, going AI native. But it didn’t take long to realize:
Becoming AI-native isn’t about adding AI features. It’s about rethinking everything—how we design, build, ship, and learn. We’re convinced It’s going to shift, and not gently, the leaderboard in every industry.
Lesson 1: The Rapid Evolution of AI Isn’t Just Hype – It’s Operationally Disruptive
Our early AI experiments were exciting and eye-opening—especially within one of our flagship products, a leading game-based workforce training platform. We built rapid-fire prototypes for AI-generated game logic and narrative content. What we thought would be a straightforward path to productization became a whirlwind.
New model releases would change what was possible (and break what already worked). Users’ expectations changed week-to-week. Our dev roadmap kept shifting. We had to dedicate R&D roles just to stay ahead.
This isn’t unique to us. A Deloitte study found that successful AI integration is less about speed and more about organizational flexibility. Building the infrastructure to respond quickly—not just ship quickly—is key.
Lesson 2: UX Is No Longer Static – It’s a Moving Target
One of the first things AI broke was our UI. Interfaces we once considered clean and intuitive suddenly felt clunky next to ChatGPT. Users wanted more fluidity, natural language, and contextual memory.
We redesigned one AI tool’s UX four times in eight weeks. Each cycle was driven by user feedback and data. The takeaway? Iterative design isn’t a best practice anymore—it’s a survival tactic.
This aligns with insights from Salesforce UX researchers, who argue that modern AI products must blend structured flows with open-ended conversational systems.
Lesson 3: Your SDLC Will Break – Plan for It
Every part of our SDLC had to be rethought. Testing deterministic software is straightforward. But testing AI? Not so much.
Our test cases began failing in real-world conditions. Prompts degraded over time. Outputs changed between runs. We adopted tools like promptfoo and LangSmith to evaluate output quality. We built prompt regression testing pipelines. We now use LaunchDarkly to progressively roll out changes in a controlled way.
Various industry studies seem to support this as well. McKinsey recently noted that AI forces a reimagining of product development pipelines—those that adapt fastest will innovate fastest. This square with our experience as well.
Lesson 4: Traditional Team Structures Slow You Down
AI-native development exposed deep friction in our team setup. Silos between designers, product managers, and engineers caused delay. Decisions took too long. Handoffs introduced errors.
We restructured around smaller, cross-functional pods that worked closely from ideation to shipping. The result? Tighter feedback loops, faster iteration, and more accountability.
Broader industry trends attest to what we see first hand. Skillsoft reports that 81% of executives acknowledge major AI skill gaps. Solving this requires not just training but redesigning team structures to support learning and experimentation.
Lesson 5: AI^NI Is the Future – Hybrid Intelligence Wins
We coined the term AI^NI (Artificial Intelligence raised by Natural Intelligence) to describe the model we believe in. It's not just AI automation. It's not just human expertise. It's the synergy of the two—and it’s what powers the future of product development.
Whether it's:
- Developers using AI coding assistants to 10x their productivity
- Product managers collaborating with AI copilots to map roadmaps
- Designers using AI to test interaction models
—it’s the human + machine feedback loop that delivers speed, scale, and precision.
CursorAI’s rise proves this. By building an IDE from scratch around agentic AI, they didn’t just iterate—they redefined how developers work. We’re seeing similar transformation in legal (HarveyAI), customer support (Forethought), and enterprise operations (Adept).
Lesson 6: Engineering Still Matters – Why Coding != Vibing
There’s growing hype that AI can replace traditional programming with natural language. But let’s be clear: engineering still matters. "Vibe coding"—using AI to draft and assemble code through natural conversation—can be an exciting way to get started. It’s fantastic for prototyping, brainstorming, and accelerating early-stage development. But it becomes counterproductive quickly when you’re building real, reliable products at scale.
Code isn’t just a way to instruct machines—it’s a way to clarify thinking. As Dijkstra once argued, the act of formalizing thought through code forces a level of precision that natural language can’t match.
While LLMs are useful for prototyping, debugging, and accelerating mundane tasks, they can’t replace the architectural rigor, logical precision, or design clarity that comes from skilled engineering.
We’ve found the best results when we combine LLMs with structured development processes: natural language as input, code as the final spec. LLMs make great collaborators, not replacements. The process of translating thought into code remains a core part of building scalable, resilient systems.
The Bottom Line: Transform or Be Left Behind
Going AI-native isn’t easy. We feel It’s very hard but necessary. As Ravi Gupta said, “Winning in this new world requires a full ‘re-founding’ of your company.” We are re-founding, and helping others re-found their companies and take advantage of the exciting possibilities.