Vibe coding is an extension of developers ability to tinker using LLMs, but its true power will come in enabling adjacent personas like product managers to more closely collaborate with their dev teams.
Tinkering is indeed in important. In the heady days of 2023 when the exploration of LLM tooling in the development workflow was still quite nascent, my colleague Kate Holterhoff wrote a piece entitled AI is for Tinkerers. In this piece she talks about how developer communities who like to kick the tires on tooling play a significant role in advancing technologies and moving them forward. “If it shows promise developers will pry it open, test its limits, and iterate on it.”
In the subsequent years, LLMs and AI code assistants have evolved from toys to tools. They are widely deployed in development workflows. According to the 2024 DORA report:
At the individual level, we found that 75.9% of respondents are relying, at least in part, on AI in one or more of their daily professional responsibilities.
Chatbots were the most common interface through which respondents interacted with AI in their daily work (78.2%), followed by external web interfaces (73.9%), and AI tools embedded within their IDEs (72.9%).
We continue to see a range of reactions to AI. There are people and companies that have seen the value of the tools as the path to the future.; there are those who see promise but are caution; there are people who remain skeptical and threatened.
There are concerns about how AI impacts the entirety of the stability of the entirely SDLC not just the generation of code. (These concerns are in fact also raised in the same 2024 DORA report cited above. The report has some counterintuitive findings about system stability and throughput, particularly as they relate to AI. I did an analysis on the DORA report findings if you’d like to know more: DORA Report 2024 – A Look at Throughput and Stability)
Even with this range of reactions and continued reservations, the adoption metrics for AI code assistants indicate we’ve moved beyond simple tinkering. That said, the tinkerer spirit is alive and well, though in 2025 it now takes the form of “vibe coding.”
Vibe coding is an extension of developers ability to tinker using LLMs. What is notable to me, however, is not developer vibe coding. What I find interesting is how this enables a wider constituency of makers.
For years no-code and low-code platforms have promised to enable “citizen developers,” allowing non-technical personas to create applications. While this is sometimes successful, new platforms require procurement and licenses, training on how to successfully use the interfaces, and cross-departmental effort around platform adoption.
As I pointed out in 2023, these platforms have been on a collision course with LLMs. Tools like Cursor and GitHub Copilot are now in a place where: natural language lowers the tool’s learning curve for non-technical users, which means that now non-technical users can interface with their development counterparts in the developers’ tool of choice. This is immensely powerful for adoption for all parties.
So in the spirit of AI is for tinkerers, vibe coding is for your product team.
This excellent post from Nathan Boden, CTO of CXone Suite at NICE that captures the sentiment well:
Tools like Cursor are opening up a new dimension in how we think about building software. For the non-technical crowd, you can “vibe code” using nothing but prompts and quickly create working prototypes.
For Product Managers, this means rethinking how we define “requirements.” We need to move away from statically written use cases with acceptance criteria and toward dynamic collaboration that rapidly translates intent into working software.
If you are a product manager, your ability to define a requirement is no longer confined to written documents and annotated designs from your UX team. You have the ability to prototype on your own. You can iterate on ideas yourself before you get the development team involved, and you can come to meetings with ideas that are ready for collaboration.
The earlier we as teams iterate and experiment, the cheaper and easier it is to do so. As Karpathy says, with vibe coding you can easily try things like “decrease the padding on the sidebar by half.” Sometimes these are merely aesthetic changes, but sometimes asking for these design and UX changes late in the development process can have unintended flow and experience impacts that go beyond just a “look and feel” change. PMs that are empowered to build and experiment is another route to unlocking development velocity.
The way PMs express their requirements is going to evolve, and the PMs who can use AI code assistant tools well are going to have an advantage over those who continue to rely on static text and design documents to communicate their user needs and acceptance criteria.
Related research:
- DORA Report 2024 – A Look at Throughput and Stability
- AI is for Tinkerers
- AI disruption – code editors are up for grabs
- The Pending Collision of LLMs and No-Code/Low-Code Platforms
Disclosure: GitHub is a RedMonk client.
No Comments