2 days Ago
ISTANBUL, TURKEY - MAY 06: A woman views historical documents and photographs displayed in a high ...
More tech art installation at Salt Galata on May 6, 2017 in Istanbul, Turkey. The "Archive Dreaming" installation by artist Refik Anadol uses artificial intelligence to visualize nearly 2 million historical Ottoman documents and photographs from the SALT Research Archive. Controlled by a single tablet in the center of a mirrored room the artist used machine learning algorithms to combine historical documents, art, graphics and photographs to create an immersive installation allowing people to scroll, read and explore the archives.
The SALT Galata archives include around 1.7 million documents ranging from the late-Ottoman era to the present day. The exhibition is on show at SALT Galata art space through till June 11, 2017.
(Photo by Chris McGrath/Getty Images) AI is fundamentally changing how software is built. Gone are the days of facing an empty editor, writing each line of code individually. While AI copilots are now common, with many powerful tools providing such functionality, and models like ChatGPT providing increasingly capable code, what is next? Enter Vibe Coding.
How AI Has Changed Code AI has had a fundamental impact on coding - simply by being able to write code from a text prompt. Prior to AI, code generation tools were largely structured and domain-specific. For example - in AI itself, one could specify a sequence of operations and have AI string together the prewritten modules that execute the sequence, adding appropriate configurations and links.
Now, with just a text question, AIs like ChatGPT can select appropriate open-source modules, figure out how to link them together and present you with ready-to-run code and detailed explanations. The functionality of the code is also improving with every new release of these AIs. This ability (now also present in coding tools as a copilot function) has downstream implications for coding skills, education, and software development practices, generating such trends as prompt-driven programming , and N-of-1 Programming .
What Vibe Coding Adds Vibe coding, a term coined by Andrej Karpathy , drives this trend further. In particular - Vibe Coding focuses on the “vibe” of the software, rather than the code. The concept of using English (written or spoken) as a way to specify what code needs to be written remains, but the focus is now shifted to the overall product being created, and the role of the human in designing and guiding that through a series of conceptual directions to an AI, which result in the AI iteratively refining, fixing and adding functionality to the code.
An important shift, as noted in Andrej Karpathy’s original message on X , is that humans may not even read these code versions - but rather assess them by their behavior on tests run by the AI at the human’s behest. The observation also notes that the code can grow beyond the human’s comprehension very quickly. Can You Give Me An Example? Here is the process I used to generate a simulator for a chemical reaction.
I had a week’s worth of experimental data on a chemical process. I also knew the name of the chemical process that occurred. I provided this information to an AI.
It was able to find the correct mathematical equations to describe the process and use those equations to build a simulator in Python. In the first iteration, the simulation did not capture the behavior in the experimental data well. I explored (using the AI) several real-world complications that may have affected the outcome.
After several iterations, I leveraged my domain knowledge to select a few commonly occurring complications and asked that these be added to the simulation. This process continued (10-15 iterations) until I felt that the simulation code adequately captured the complexities of the physical experiment, and the simulation results reflected an increasing closeness to the experimental. In each case, the AI did the work of generating and testing the code, also providing concept suggestions upon request.
I was the expert guide, selecting between options and making final decisions. I read the code every few iterations, but not every time. What It Means for Business Whether vibe coding is the name we will give to the future of this trend is unclear, but it is clear the overall trend is here to stay.
Y Combinator reports that over 25% of its startups are now relying on AI for 95% of their code base , while Google recently reported that about 25% of its new code is AI-generated . The trend has important implications for business, raising such questions as: - Should a business be comfortable putting into production code that no human employee has ever understood? In many cases, complex code bases survive far longer at companies than the people who wrote them - so it is not exactly uncommon to have code that no one on the premises understands. However, such code usually still exists because it has proven itself over time.
The idea of new code that no one understands is, well, new. - Super speed prototyping? Perhaps the best initial value/risk compromise is using this type of development for extreme-speed prototyping. Perhaps we have reached the stage where product development is not initiated just by business cases but also by quick prototypes validated by customer interaction that exist in nearly full-fledged form before production development even begins.
- What does it mean for the volume of code? Vibe coding and related trends all mean that the volume of code (to be stored, versioned, and protected in an IP context for business) is growing dramatically. Businesses will need to factor this into their software development practices. - What about testing, quality, security, etc.
? Production code requirements will not change because AI generates much of the code. However, how much can the developer now contribute to security and test reviews, when they can only speak to the what, but not the how, of the code? - What do software engineering teams of the future look like? It is clear that these trends benefit senior engineers, who have the knowledge to select what to build and enough instincts to detect trouble in tests even without reading the code. It does however beg the question of how future engineers will reach this knowledge level to vibe-code effectively if our learning paths (writing small bits of code, tests of algorithmic correctness) are no longer part of their job description.
Personally, I do not believe this trend spells the end of software engineers (although the question has been asked). I do however believe it signals a shift from coding to software development , where the overall product (the what), and the behavior of the complex system, take precedence over who/what wrote a given piece of code. What Should Your Business Do? A few things your business should do - Ensure that your software development managers are regularly upskilled in new trends such as vibe coding.
- Encourage discussion about how these approaches should be rolled out to teams, including guidelines for handling the issues above in ways that are suitable for your business. Takeaway Trends like vibe coding have significant potential to increase the speed of software development and the volume of customer-facing features developed, making it critical for businesses to follow and adapt to remain competitive. The guidelines above can help you “surf the wave” of AI-powered software development practically and safely.
.
