Vibe Coding… Time for a New Zeitgeist?

The more I read about vibe coding, the more I think about its implications. The world already deals with more noise than signal, more heat than light, with the increasing quantity of unmoderated AI-generated content on the web. With vibe coding on the rise, what will that do to the digital world? This article (paywall) in the FT consolidated a couple of notions I’ve been noodling on for a while.

The Wonders of Modern Photography

I’m reminded of the shift from physical film to digital cameras (and then, to digital cameras in everyone’s back pockets!). With physical film, you really had to consider every shot you took: the signal-to-noise ratio mattered! The cost and physical volume of film stock, together with the time to develop, review, and store all the physical images, meant taking care with each one of those images.

Then along came digital. Suddenly, we could all take hundreds, even thousands of photos—and as soon as high-quality cameras made it into our phones, millions of us could take ‘professional-quality’ pictures. The barriers to entry came tumbling down: photography was popularized!

This accessibility is a wonderful thing: it empowered lots of great photographers and artists who otherwise would never have had the opportunity. But it also means that the sheer number of ‘bad’—poor quality, ill-considered, bland—images has increased exponentially. And at some point, the overall cost to us of the ‘noise’ outweighs the value of the ‘signal’.

Is this to suggest that we should all stop taking photos? No! Rather, consider that although the tools allow you to produce more, and to produce it more quickly, that’s not the limit of their capability. Maybe those other capabilities are worth exploring.

The Democratization of Technology

AI-supported software development opens up tremendous opportunities to take great ideas—innovations that make our work easier, our lives better—and get them into people’s hands quicker. But just because we can build something, doesn’t mean we should. And being able to deliver product to market at record speed isn’t always the right move.

We’re seeing this phenomenon already in other creative realms impacted by Generative AI. “AI Slop”, “Digital Clutter”, “Filler Content”… it all leads to the same end result: signal-to-noise ratio degradation. The consequence is that we have to spend more time—not less—filtering, evaluating, or verifying material to make sure it’s worth our time to engage with it. So with software development. The promise of development democratization is real; so is the risk of “software slop”.

Asking the Right Questions

Here, then, is a thought: instead of simply shortening the delivery cycle, let’s instead repurpose the time we are going to save in reification to do two things. First, ask the question, ‘Will what I’m building really make the world a better place’. Ask honestly. Seek third-party validation. If the answer is ‘no’, then it’s OK to stop: you will have other ideas!

If the answer is ‘yes’, then take the time to build something that truly meets the requirements of the audience. Well-structured and comprehensive requirements gathering has been rather forgotten in the age of minimum viable product, as has painstaking (human-driven) software testing. Let’s repurpose the hours we’re going to save on the engineering side to make sure that the products and services we’re building are fit for purpose.

Getting there from here will take some thoughtful leadership and a change in perspective. Executive management will need to take a step back and allow teams time to adjust. We will need to measure progress differently. “How many lines of code did we deliver today?” might become “How many requirements did we validate?” Boards need to dig into the process: “How did we make sure this product meets the needs of our customers? To whom did we talk? Why?”. And engineers will need to embrace new skill sets to meet a different set of challenges: reviewing code (critically, in depth, and with skepticism), holistically understanding and developing requirements, and ensuring that the product actually meets those requirements.

These adjustments will not be easy. We’ve become inured to things operating at ever-increasing pace, with ever-increasing volume. And learning new skills—building new curricula and professional development pathways—will take time.

But if we end up building better solutions that actually meet real needs, and meet them well, it will have been worth it.

If you’re interested in other thoughts I have on digital identity, privacy, and corporate governance, I encourage you to read through this site or follow me on LinkedIn .