The new era of PM coding is not about vibes

Andrej Karpathy coined "vibe coding" in February 2025 as a half-joke. He described it as building software by prompting AI, accepting whatever code comes back, and deploying without really understanding how it works or whether it solves the actual problem. The thinking was that you could just keep iterating until you got it right. The term caught on fast — by last summer, Collins Dictionary named it Word of the Year. That must have felt like an honor.

The term resonated because it captured something true. AI has made it possible for just about anyone to ship software, and it feels like the good kind of magic. The following is also true: You can build prototypes and code features just because you can, and not because you should.

Building with AI is now much faster than figuring out whether you should build in the first place.

I have recently heard this cautionary story from enterprise product teams. And you can increasingly find similar stories online. The pace of prototyping and developing with AI feels exhilarating at first. Until you realize you built something no one actually wants. In those cases, generating code is outrunning the need for it.

The acceleration itself is undeniable. AI assistants write code in minutes that used to take days. My co-founder, Chris, and I have been tracking this, and we predicted last year that the percentage of code written by AI will quintuple year over year from here on out. We call this the Waters theory.

Given the power of AI and what is now possible, product builders who want to create real value will need to stay disciplined. The approach is not so different from what the best teams use today — it is the same proven fundamentals applied with new capabilities.

This disciplined approach is what will define PM coding. And it is what product managers already know how to do.

Vibe coding vs. PM coding

Vibe coding is what happens when you just go. Prompt an AI model, get something working, and ship quickly so you can see it in the real world.

PM coding takes a more disciplined approach that is always needed to build something lovable. Start by understanding the customer problem deeply, then use AI as a force multiplier for execution once you are clear on what needs to happen and why. The goal is to be fast and "plan-full" rather than fast alone.

AI amplifies the work. The discipline is what makes the amplification lead to creating a Minimum Lovable Product — a concept we pioneered in 2013.

PMs will be the new builders

PM coding works because product managers hold deep knowledge about users — problems and thoughtful ways to solve them. That context makes PM coding effective for any product work, but it should be particularly powerful for building business applications and internal tools. (Product managers are not the only ones who hold this kind of context. People across marketing, sales, support, and operations teams also have deep user insights and see how to build a better future. They can apply this same disciplined approach.)

I have recently talked to product teams who needed a partner onboarding portal, but could never get it prioritized. Others wanted a product catalog for sales teams that always sat at the bottom of the backlog. Some needed dashboards surfacing real-time field performance data or event-tracking systems that captured what spreadsheets could not. You probably have your own version of this list.

The reason these applications languish is not that they lack value. Someone in the organization, often the PM, knows exactly what is needed and why it matters. What has always been missing is the capability to build them without engineering resources. If PM coding changes that, the implications are significant. Product managers could bring both the knowledge and the execution. Applications that solve real problems for colleagues could get built without pulling engineering away from existing revenue-generating applications.

Where this leads us

The era of PM coding is about closing the distance between knowing what needs to happen and making it happen. It empowers the knowers to also be the doers. Doing no longer requires precious engineering resources that are hard to galvanize for applications that do not generate revenue. This is why PM coding will initially thrive where software does not already exist. Applications will quickly be built to serve colleagues who will be more productive with new tools — colleagues who could not justify the cost of building or buying software in the past.

The era of PM coding is now here.

The combination of disciplined product fundamentals and AI execution is more powerful than I think most people realize. Product managers who have always known what needed to be built but lacked the technical skills or technical resources to build it are about to have those constraints removed.

Stay connected and let's continue to explore the future of product management — subscribe to the Aha! newsletter.