Coined by AI researcher Andrej Karpathy in early 2025, vibe coding describes a new software development paradigm where, instead of writing code line by line, you simple tell a large language model (LLM) what you want in natural language and let the AI generate, execute, and refine the code mostly on its own. 

In other words, even someone non-technical like me can now build things. 

Karpathy described it (in his X post) as doing things like: “You fully give in to the vibes, embrace exponentials, and forget that the code even exists.”

How vibe coding works

In practical terms, this is how vibe coding works:

1. Prompt-driven design: The user’s prompt is the core. How you ask matters: Be specific, clear, use examples, and constraints.

2. Iterative feedback loop: You test what’s generated, see what works or fails, then refine the prompt or give corrective instructions. Similar to “trial and error” but with AI as the heavy lifter (eg “make X faster,” “move Y up the page”)

3. Lowering the barrier to entry: Non-expert users can build working prototypes or apps without deep programming knowledge. The AI bridges a lot of the technical detail, making this available to pretty much anyone.

4. Trade-offs and robustness: Because you’re not always inspecting every line of code, there are risks: security vulnerabilities, technical debt, maintainability, edge-cases that get missed. Also issues when scale or production environments call for structure, reliability, testing, etc. This isn’t a final product (yet), this is a prototype.

Why you should care

Here are a few reasons you should take this seriously and encourage individuals and teams to experiment (but not get too distracted!):

• Faster prototype and feature development: Want to spin up a localisation widget, experiment with article recommendation, or build side projects or internal tools (like a data dashboard)? Vibe coding can dramatically cut the time to the first working version.

• Empowering non-engineering teams: Product managers, designers, data analysts might build their own tools or mockups without going through full engineering cycles. This could unlock more agile experimentation and autonomy for people to show mini versions of their idea rather than explaining it trying to get buy-in to build.

• Cost and resource shifts: If you lean on AI for much of the “writing” of code, the role of engineers shifts: maintaining, hardening, reviewing, and ensuring AI-generated work meets quality, security, and performance standards.

Also beware there is a risk of tech debt and maintenance burden: If foundations aren’t solid, they may require serious “cleanup” later. Because code is being generated without full human oversight, subtle bugs, security holes, and inefficiencies are common. 

If you assume the AI’s output is good without sufficient testing, things can go wrong — especially in production. Therefore there are governance, security, and compliance implications. When non-technical teams build stuff, oversight is critical. Data privacy, authentication, access control, proprietary and sensitive content need to be managed.

Culturally speaking you also need to beware of over reliance, or false confidence. Because you’re often removed from the mechanics of code, you need to develop the ability to test, monitor, and verify without being a full code expert. It can also be distracting as things can be spun up too easily.

Everyone loves innovation, but you don’t want everyone doing it all the time. 

Specific uses for news organisations

Given all that, here are some thoughts on how news organisations might navigate vibe coding:

1. Product: testing new features (apps, newsletters, paywall tweaks, rec engines).

2. Technology and data: internal dashboards, automation, rapid backend experiments.

3. Editorial: support tools for journalists, story tagging, summarisation, fact-checking workflows.

4. Commercial: campaign landing pages, ad product prototypes, even event microsites.

5. Organisation-wide: lowering barriers to innovation, but needing governance to avoid “shadow IT” chaos.

In summary, Vibe coding feels like a tipping point in how we think about software creation. It offers media companies the promise of faster innovation, more empowerment for non-technical staff, and surprising creativity. But it comes wrapped with trade-offs: security, maintainability, quality, and governance are very real issues. 

For the bold, it’s a chance to rethink who builds what and how. But it’s not a silver bullet. Like all tools, its value depends on how well you use it, how well you monitor its risks, and how much you’re willing to invest in cleanup when the vibes aren’t perfect.

If you’d like to subscribe to my bi-weekly newsletter, INMA members can do so here.

Banner art by Adobe Stock DIgilife.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .