For years, I've described good product development as a delicate balance between three fundamental perspectives:
- Business people answer how we pay for it and how we make money from it—whether through revenue, cost reduction, or strategic advantage
- Engineers tackle how we build it and how we maintain it over time
- Product people focus on what needs it solves and who has those needs
The magic happens when all three perspectives are balanced. The most elegantly designed product remains worthless if you can't afford to build it or maintain it over time.
I've held senior positions across all three domains throughout my career, but never believed one person could effectively perform all three roles simultaneously. The analytical mindset needed for debugging code conflicts with the empathetic thinking required for user research. The pragmatic constraints of engineering budgets clash with the ambitious vision needed for breakthrough products.
This limitation held true until recently.
Over the past week, I generated roughly 30,000 lines of code alongside another 20,000 lines of documentation using Claude Code. This felt completely different from traditional coding. Rather than wrestling with syntax and library documentation, I found myself pair programming with an AI agent, conducting real-time code reviews, and focusing on architectural decisions.
For the first time in my career, I can envision a path where experienced professionals might effectively operate across all three domains. AI allows individuals to offload the mechanical work while focusing on strategic decisions. This enables them to produce diverse artifacts traditionally siloed across different subject matter experts.
Of Machines and Men: The Reality of AI Coding Partners
Working with agentic code generation feels remarkably like managing Lennie from Steinbeck's Of Mice and Men. The AI possesses the raw power of two or three experienced developers, but left unsupervised, it will crush the bunny every time.
When debugging a failing test, the agent would attempt a fix once or twice, then decide the test simply needed to be disabled or "simplified" beyond recognition. A gentle redirect—"No, we need to fix or rewrite the test appropriately rather than disabling it"—would usually produce the right solution. The AI wants to solve every problem by adding more content, but programming often requires subtraction and restraint.
"It's not the notes you play, it's the notes you don't play." — Miles Davis
When building baseline architecture, it reinvents wheels with enthusiastic inefficiency. At one point, it correctly identified the need for logging in a function, then proceeded to write a custom Logger class from scratch. When I questioned why we weren't using a battle-tested logging package, it immediately agreed and refactored to implement a proper third-party solution. But I had to recognize the misstep and suggest the alternative.
This dynamic mirrors working with a talented but inexperienced engineer—one who lacks pattern recognition and doesn't know what tools already exist. To get production-quality results, you need serious engineering knowledge to guide the process effectively. George must understand the complex jobs on the farm to get the most out of Lennie.
The key is constant supervision with small, specific tasks. Broader, more ambiguous requests often lead to cycles of refactoring and questions like "Why aren't we using our common logging pattern here?" The available context window is fixed, so working with an AI is like working with someone who can't form new long-term memories—you're constantly reminding it of established patterns and decisions.
"Tell me about the rabbits, George." — Lennie
The Convergence Opportunity
For product managers and designers who possess these engineering fundamentals, the possibilities are transformative:
- Design systems scale efficiently — I can generate design system and UI code rivaling the output of most front-end teams, freeing engineering resources for higher-value problems
- Prototyping becomes table stakes — I can rapidly create high-fidelity, functional prototypes instead of static mockups, replacing debates and hypotheticals with testable interactions
- Data exploration democratizes — I can analyze datasets far more effectively than traditional BI tools allow, without needing to master R or Python
Thinking realistically, this opportunity has a crucial limitation: it requires substantial technical knowledge that most product managers and designers currently lack. The combination of business acumen, product intuition, and technical depth needed to effectively supervise AI code generation remains relatively uncommon—even among talented people I've hired and would hire again.
The Succession Challenge
The rise of AI-powered development creates a succession challenge: we're raising the bar for entry-level professionals while making their traditional learning paths obsolete. Companies need to invest in new apprenticeship models that develop systems thinking alongside technical skills.
Rediscovering Joy in Building
On a personal note, I haven't genuinely enjoyed writing code for 10-15 years. I avoided coding because constant syntax lookup killed my creative flow. Knowing what I wanted to accomplish but spending hours researching specific classes and methods was soul-crushing.
AI coding agents have changed this completely. I can focus on architectural and design decisions—the parts I actually enjoy—while the AI handles syntactic details. Building software has become fun and creative again instead of frustrating and slow.
"When you ask creative people how they did something, they feel a little guilty because they didn't really do it, they just saw something." — Steve Jobs
The ability to focus on creative problem-solving rather than mechanical implementation suggests we're entering an era where role boundaries become more fluid, at least for those with the right foundation. But it also raises questions about code review processes, quality assurance, and team structures that we're only beginning to explore.
Looking Forward
This isn't theoretical - the convergence is happening now, but only for professionals who bring together business acumen, product intuition, and technical depth that remains relatively rare. Companies that recognize this and invest in developing these hybrid capabilities—while also creating pathways for the next generation—will have significant competitive advantages.
In my upcoming posts, I'll dive deeper into the practical realities of working with AI coding agents, explore specific opportunities for product and design roles, and examine potential solutions to the training crisis we're creating. I'm also beginning an experiment to build an enterprise-class design system using these methodologies, which should provide concrete examples of what's possible.
The question isn't whether AI will change product development—it already has. The question is whether we'll adapt our roles, our teams, and our training to harness this change effectively.
This is the first in a series exploring how AI coding agents are reshaping product development. Follow along as I document building a complete design system using these tools, starting soon.
Top comments (0)