I’m noticing a pattern: most founders now approach me with a vibe-coded product. It’s normally great at communicating their vision. But I’m not yet seeing non-technical founders able to turn it into a great product.
What I see is that they race to 90% done and then get stuck. At the same time, our software engineers at Platform Venture Studio are busier than ever.
Given all the advances in the core models and the coding tools, why is that?
This is not intended to be a specifically optimistic or pessimistic take. Things are changing quickly. It’s possible that in a few weeks, Opus 6 or GPT 7 can one-shot a great product.
But I think the current gap boils down to 4 key areas where competent, product-minded software engineers still have an advantage:
1. Systems Thinking
Software engineers naturally ask, “What is the underlying abstraction?” versus repeatedly solving specific cases. They can generalize A as a special case of B and build a system to solve B.
They also tend to be good at problem decomposition; turning a big, hairy problem into a series of independently solvable subproblems.
It’s true that traditional maintainability and tech debt are less of an issue when AI is doing both the building and the maintenance. But, in their need to please the user, I’m seeing AI tools adding layer upon layer of complexity, which eventually starts to hurt overall system stability and performance.
2. Language & Terminology
Experienced software engineers have spent years turning fuzzy requirements into explicit requirements, edge cases and acceptance criteria. They are also able to converse with LLMs using established technical terms.
They may not like “Claudish,” but they are able to learn it.
3. Process & Temperament
Software development requires tinkering and relentless iteration to get to a great product.
This is a process that needs to be learned, but it’s also a mindset. It requires sustained attention and a high degree of frustration tolerance. It’s also lonely, so people who thrive on human interaction fare less well.
4. Judgment & Taste
Product is defined as much by what you choose not to build as by what you do build. You’re not done when you can’t add any more; you’re done when you can’t take any more away.
LLMs didn’t get the memo. Their natural tendency is to always want to output more tokens, and they seem to be wired to keep adding new things.
As to UX, plenty of engineers have never been good at it. AI may already be better at UX than the average human engineer. But, right now, it doesn’t seem to be as good as the best product engineers. AI is good at drawing on its enormous library of patterns for any given case but is less good at building a consistent information architecture across a whole product.
Rather than making everyone a 10x engineer, I think AI may actually be turning 10x engineers into 100x engineers.
This means that AI is perhaps increasing the productivity gap between great engineers and mediocre engineers, rather than flattening it.
Founder @ TruAnon — distributed identity confirmation platform
that first 90% is hard, but it's the last 90% that kills us! -- seriously, I have noticed no real difference in actual end quality just yet - a non-technical person apparently does not know what to ask, when to push back and when to pursue patterns and bugs in a way AI can complete. So, my results are not the same as a non-technical "engineer" vibe coding. In other words, it does not make you a great engineer, it's just a lot of people find the ability to speak in code seems magic, for those who already do speak in code? we're more skeptical of the results and guide a product to quality the same capacity we crafted as lifelong engineering experts. AI does not give you this capacity, do not replace your architects and most experienced with folks who vibe. They cannot achieve professional results even with a tool that seems magic. We leverage magic in a different way.