Vibe coding is thought by most to replace coding, but it could be replacing something completely different — product management.
In a recent post, entrepreneur and investor Naval Ravikant reframed the conversation around AI-assisted development with a provocative claim: “Vibe coding is the new product management. Training and tuning models is the new coding.”
While much of the tech industry debates whether AI will replace programmers, Naval suggests we may be focusing on the wrong question. His thesis contains two interconnected predictions about how AI reshapes software development roles.
The Displacement of Product Management
Naval’s first observation is that “vibe coding” — using natural language to describe what you want built — fundamentally performs the same function as traditional product management. When developers articulate user needs, define features, and specify functionality through conversational prompts, they are doing the work that product managers have historically done.
Traditional product managers serve as translators between business requirements and technical implementation. They write specifications, create wireframes, prioritize features, and communicate the vision while engineers handle execution. But when a developer can describe the desired user experience in plain language and watch AI translate it directly into working code, that translation layer compresses significantly.
The “vibe” becomes directly executable. The person with product vision can potentially build it themselves, without needing an intermediary to document requirements or manage backlogs. This doesn’t eliminate the need for product thinking, but it raises questions about whether it needs to exist as a separate specialized role. This already appears to be happening in the real world — LinkedIn recently combined Product Manager, Frontend and Backend Engineering and Designer roles into a single “Full-stack Builder” role that performs all these functions.
The Evolution of Technical Work
Naval’s second claim addresses what engineering actually becomes in an AI-native environment. If AI handles routine coding tasks, the valuable technical work shifts to model operation: selecting appropriate base models, curating training data, fine-tuning for specific use cases, and iteratively improving model outputs.
This represents a fundamental change in required skills. Traditional software development rewarded algorithmic thinking, logical precision, and deep knowledge of data structures. Model work requires statistical intuition, experimental design capabilities, and the ability to diagnose unexpected model behavior. The work becomes less deterministic and more empirical. Interestingly, much of this work is also being automated — Anthropic engineers say that Claude Code wrote most of the code for Claude Code, and OpenAI’s employees have said the same thing about Codex.
Implications and Limitations
If Naval’s analysis proves accurate, several organizational implications follow. Developers who combine technical skills with strong product intuition become increasingly valuable. Product managers who cannot or will not work directly with AI tools may find their role diminishing. Teams could become smaller and flatter as individuals handle both vision and implementation.
However, several factors complicate this transformation. Vibe coding currently works well for conventional applications but may struggle with truly novel or architecturally complex systems. The strategic elements of product management — market research, competitive analysis, stakeholder coordination — don’t disappear simply because implementation becomes easier. And current AI models still require substantial human oversight and correction.
Historical Pattern and New Territory
Naval’s observation fits a recognizable pattern in technology evolution. Each new abstraction layer — from assembly language to high-level languages to no-code platforms — prompted predictions that programmers would become obsolete. Instead, what typically occurred was a shift in what programming meant and who could participate in it. Naval suggests we’re witnessing the next iteration of this pattern, with a significant difference: this time, it’s not just implementation being abstracted, but also the product specification layer itself. The entire stack of software creation is potentially reorganizing.
And Naval’s framework raises a final consideration: if vibe coding replaces product management and model tuning replaces traditional coding, what happens when AI becomes proficient at model tuning as well? The answer may determine whether we’re simply ascending one level of abstraction or approaching a fundamentally different regime where human contribution centers primarily on vision and judgment rather than specification or implementation.