By Paul Maker
I’m a firm believer that — like oil and water — vibes and coding don’t mix particularly well.
When we code, we’re following rules and concepts to make sure programs are built on proper foundations. In contrast, vibes are about intuition and what feels right. Mash them together and you will inevitably end up with inconsistent software with inherent reliability issues.
In the short-term, vibe coding is an approach that creates confusion and buggy problems. But in the long-term, the stakes are far higher. This potent mix is a recipe that could lead to model collapse.
The long-term risks of sloppy code
Vibe coding replaces experience with vision. And while it can be a great way of experimenting and opening up the magic of software development to more people, when used to create code that will actually power systems and be relied on by others, it’s not fit for purpose.
Products built on AI-generated code that hasn’t been properly stress-tested weaken the integrity of everything built on top of them. Every time we lean on AI-generated code, or train new models on the outputs of vibe-coded ones, we are polluting the data sets and weakening the infrastructure that forms the foundation of software.
It’s a feedback loop where slop feeds off slop. The code gets fuzzier, the bugs more frequent, the quality of results steadily worse. Ultimately, the very models themselves will be unable to tell fact from fiction — rendering results and outputs useless.
This is model collapse. Constant hallucinations and errors, faulty systems that can’t be fixed without a complete rewrite, and a breakdown of trust among customers.
When we rely too heavily on approaches that cut corners — skipping testing, overlooking governance and avoiding hard questions — we won’t just get bad apps, we’ll break the foundations that models rely on.
How to build better
It’s time to re-establish some fundamentals. To avoid model collapse, here’s what we can do:
Double-down on data governance. If you don’t know where your data came from, or whether you can trust it, you’re essentially building on sand.
Build the infrastructure needed to accurately classify and label documents, before enriching them with top-quality metadata. Then ensure this data is stored correctly, backed up, and given the right security and permissions to enable good governance.
If you’re keen to optimize delivery, there are AI data governance tools that can do it for you. Establishing solid governance gives you greater control over the AI systems that work on your data. (Oh, and just another reminder to always have a backup.)
Train engineering muscle
While AI can speed up development and enable teams to ship code faster, leaning too hard on it can cause core skills to wane. It’s vital to train developers properly and continue to invest in junior talent. All colleagues need to understand what “good” code looks like, so they can ensure the quality remains high and become the senior leaders of tomorrow.
Building coding skills without shortcuts should be a focus for junior developers to ensure that the next generation of tech talent understands what they are building at a foundational level, before they reach for optimization tools.
Aim for real-world feedback
It’s not enough to test your products in a sandbox and call it a day. An important part of the development cycle is testing software in the real context it’ll exist in. For example, beta testing your products on users might take a lot of time, but it will stress-test systems in ways that developers might never have anticipated.
In other words: don’t rush feedback. Spending time observing and measuring how your products perform in the real world means you can catch and resolve quality issues early and build reliability into a final product.
Discipline means stability
The next wave of startups have a choice: build with discipline or build on slop. One scales trust. The other scales technical debt. By resisting the temptation to cut corners, we can ensure strong foundations and avoid model collapse.
Paul Maker is chief technology officer at AI company Aiimi, whose tech and services help teams find, make sense of and retain control over their data. He leads Aiimi’s research and development on new and emerging technologies, with a particular focus on AI. When Maker is not at his computer, you will find him either at the gym or walking his dog.
Illustration: Dom Guzman
Stay up to date with recent funding rounds, acquisitions, and more with the
Crunchbase Daily.
