Using AI to write code, often without human review, has become known as ‘vibe coding’. It’s showing the glimpses of a new paradigm: experienced developers are discovering that vibe coding can massively increase their productivity, and can be used to write entire stacks in an afternoon rather than weeks. Currently, the code produced is far from perfect and there’s a lot for the developer to improve, but the promise is there.
It can be easy to dismiss the value of vibe coding, as the benefits will depend upon the seniority and experience of the person who is ‘vibe coding’. For example, someone with little or no technical experience tends to find it creates code they never could, but it’s often unreliable, hard to maintain, or has critical issues.
Experienced coders who pride themselves on effective and elegant code observe that the code AI produces is way below the quality they require. Managers responsible for development teams’ efforts also report concerns about the quality and the maintainability of code. Who wants to reverse engineer thousands of lines of AI generated code to realise it has an off-by-one error? And how will it affect the career pipeline for developers?
However, people who are experienced coders (or experienced in assigning tasks to more junior staff) are already telling me that they are seeing huge increases in their productivity. I recently spoke with a startup that had received a SaaS renewal quote that was twice the current price, as an increase in user numbers pushed them into a higher-priced tier. Instead of renewing, one of their engineering leads ‘vibe coded’ a replacement with the core functionality they needed within a couple of hours. And this trend isn’t limited to risk-tolerant startups; large organisations I talk to are now seeing developer and admin teams using AI when creating and maintaining internal tools to support their work.
Bill Gates’ frequently-cited quote from 1996 reads: “most people overestimate what they can do in one year and underestimate what they can do in 10 years”. With this in mind, let’s look at how some thought leaders have predicted how vibe coding might mature:
- Dan Shapiro’s blog ‘Spicy autocomplete to dark factory’ takes inspiration from the standards used to assess autonomous driving to assess the stages of how vibe coding might integrate in an organisation.
- Steve Yegge’s blog ‘Welcome to Gas Town‘ captured imaginations by proposing a maturity model and architecture based on a multi-agent ‘town’, and also by trying to build it.
Both of the above envisage different stages for the same core idea; vibe coding will evolve from being an aid to humans, to a point where no human has even looked at the code being run. This process will take time and won’t necessarily arrive evenly. I argue that it will spread mainly along 3 dimensions:
- how complex the service is
- how important the service is
- how risk averse the organisation is
Don’t get me wrong, there are MANY issues today with vibe coding. Social media and blogs are awash with examples of rubbish code, horrific security, ‘slopsquatting‘ and other new types of attack. But as was the case with SaaS a decade or two ago, the business benefits on the table will be too strong to resist. I strongly suspect we’ll see broad migration towards vibe coding along the above 3 axes, regardless of the security concerns.
A mitigation that’s sometimes suggested is that organisations will simply say ‘all code written by AI must be reviewed by a human’. But already that guidance is creaking at the seams for organisations that want to build aggressively. Over the next 5 years it will become increasingly common to see AI-written code in production systems that a human has never reviewed or even looked at.
