AI Is Speeding Up Iteration — And Making PM Judgment More Critical Than Ever

I. A Fundamental Question

I recently revisited Henrik Kniberg’s Making Sense of MVP, and one idea stood out more clearly than before. He uses the analogy of “skateboard → bicycle → motorcycle → car” to explain MVP. At its core, it is not about “gradually improving a product,” but something more important: validating an assumption with the smallest possible cost as quickly as possible.

In traditional product development, this kind of “upgrade path” is often extremely expensive. When I first saw these diagrams, I often thought: from motorcycle to car, almost the entire codebase would be rewritten. From an engineering perspective, it feels almost impossible. So I used to treat this as an idealized development process.

The real problem is not whether you can build a more complete version, but that the further you go, the less flexibility you have to validate your assumptions.

---

II. AI Is Changing Not Efficiency, but the Iteration Model

What truly changed my perspective is AI. I clearly feel that the rhythm of product iteration is undergoing a structural shift.

Previously, an MVP cycle looked like this:

• Design → Development → Launch
• Instrumentation → Data collection
• Analysis → Iteration

A complete loop usually took weeks or even months.

But now:

• AI can quickly generate prototypes
• Assist or even replace parts of development
• Analyze user behavior in real time
• Automatically organize and categorize user feedback

What used to take weeks can now often be completed in hours.

This is not just about being “faster.” It is a fundamental change in the way products iterate. MVP is no longer a phase-based versioning process, but a continuously running system. Products are no longer released version by version—they are constantly being tested, adjusted, and tested again.

The dream of “motorcycle → car” is becoming achievable with AI.

---

III. Agile Is Becoming an Experimental System

This also reshapes how I understand Agile. On the surface, we are still doing iterative development, but the essence has changed.

Previously, a sprint was about delivering features. Now, a sprint is closer to running experiments.

The questions I focus on are no longer:

• Is this feature finished?

But instead:

• Has this hypothesis been validated?
• What does user feedback imply?
• What should we test next?

Development is no longer the accumulation of features, but the continuous optimization of experiments.

---

IV. The PM Role Is Shifting Upstream

In this shift, my understanding of the PM role has also changed.

Previously, PM work was mainly:

• Gathering requirements
• Writing PRDs
• Coordinating design and engineering

But these are becoming less central. The real core of PM work is increasingly two things:

1. Deciding what to validate

AI can help build things quickly, but it cannot tell us:

• Which direction is correct
• Which problem is worth solving
• Which hypothesis is worth testing

If these decisions are wrong, faster iteration only means faster waste.

---

2. Extracting signal from feedback

Today’s problem is no longer lack of data, but too much data: user behavior logs, clicks, text feedback, conversations—all continuously generated.

AI can help:

• Cluster feedback
• Summarize insights
• Perform sentiment analysis

But it cannot solve a key issue:

What is noise, and what is a directional signal?

For example:

• Is a user complaint an outlier or a trend?
• Is low feature usage a design issue, or does the need not exist at all?

These judgments ultimately require human interpretation. And increasingly, this ability is becoming the core value of a PM.

From a broader perspective, I now see product development as a continuous learning system:

• AI generates and executes
• Agile provides the iteration loop
• PM defines direction and interprets signals

A product is no longer a static collection of features, but a system that continuously interacts with users and adapts over time.

---

V. Personal Reflection

My conclusion is:

AI does not make PM less important—it makes PM more important.

Because when “building things” becomes cheap, what becomes scarce is:

• What to build
• Why we build it
• What we learn from the outcome

I now interpret MVP differently. It is no longer just the “Minimum Viable Product,” but rather:

Minimum Learning Unit

Product development is no longer a 0-to-1 construction process, but a continuous experimental loop. And in this loop, what matters most is not what you build, but:

Whether you made the right judgment.