Very few phrases in software have spread as quickly, or been as misunderstood, as “vibe coding”. For some, it signals the democratisation of software development, or the practise of causal programming for throw-away applications. But to its loudest critics, vibe coding represents a shortcut culture: developers tossing four‑line prompts into an AI model, then flooding production systems with fragile code. Many engineers also dislike the term because it implies coding without really understanding the code itself or blindly accepting AI output. In turn, some see vibe coding as trivialising the real effort and expertise required in software development.
I’ve been in the industry long enough to have heard similar complaints before: a new technology trend or tool that will make developers lazy, hollow out engineering skills, and ultimately leave organisations exposed. But that framing misses the point, because it has become a catch-all to describe two very different practises, with very different implications for developer teams and enterprises.
AI-driven coding is not vibe coding
Vibe coding often distracts and undermines the very real and positive impact of adding AI to the programming workflow. The leaps in the past few months alone have meant that AI agents have rapidly transformed how software is developed. Rather than manually writing code line by line, developers can now assign complex tasks to AI agents in plain English and have them independently research solutions, write code, debug errors, and deploy systems. These agents are capable of handling long, multi-step workflows – for example, setting up infrastructure, integrating tools, testing functionality, and documenting the results – often completing in minutes what previously took days.
As a result, programming is shifting from direct coding to orchestrating and supervising AI-driven processes, where developers focus more on defining problems, guiding agents, and reviewing outcomes, rather than implementing every technical detail themselves.
This style of development still requires a comprehensive understanding of engineering principles. As a result, it should not be confused or conflated with vibe coding, where code may also get generated rapidly, but lacks proper reviews, and is deployed with limited consideration for scalability, security or ownership. When critics warn about insecure “slop” code entering production, they are often pointing to failures of process and governance. That is a legitimate concern. But it is a concern about how teams operate, not about the inherent properties of AI-assisted development.
Even long-standing sceptics are beginning to acknowledge this distinction. Figures such as Linus Torvalds, Donald Knuth, and Robert C. Martin have recognised that AI can play a constructive role when it sits within disciplined engineering practice. The conversation is shifting away from whether AI should be used at all, and towards how it should be governed.
AI accelerates existing coding practises
Teams that have always kept strong review standards, testing rigour and capable leadership will not have those practises washed away if AI enters part of their workflow. For these teams, AI helps to increase the output without lowering standards. This is because it amplifies the culture already there.
The same principle applies to security. Secure software depends on threat modelling, dependency management, access controls and continuous monitoring. If those disciplines are weak, vulnerabilities will surface regardless of whether the first draft was written by a human or generated by a model. AI changes velocity, not responsibility.
For CIOs and technology leaders, this has practical implications. Prohibiting AI tools is unlikely to succeed and may drive their use underground – and given the advances in AI-driven coding, it may even amount to a serious competitive disadvantage. The more effective response is to strengthen the controls around them. Clear coding standards, automated testing, AI-driven code review, policy-driven code scanning, and runtime observability must become the default.
As AI accelerates the pace of software development, traditional human-centric code review workflows will struggle to keep up. When code can be generated much faster than it can be manually reviewed, human reviewers quickly become the bottleneck. The emerging model is one where AI systems enforce standards, review and validate code continuously, and flag risks at machine speed, while human engineers focus on defining requirements, architecture and higher-level oversight, rather than line-by-line review. AI belongs inside a structured software development lifecycle with guardrails, traceability and accountability.
A higher baseline for good code
With these in place, AI can raise the bar for what “good” looks like in our industry. Junior developers can receive immediate feedback on idiomatic patterns, security pitfalls and performance trade-offs. Senior engineers can offload repetitive tasks and focus on architectural coherence and long-term design. Test generation, documentation updates, refactoring and coding standards enforcement can become continuous activities rather than deferred clean-up exercises.
In many organisations, technical debt accumulates because improvement work competes with delivery deadlines. AI is reducing that tension by lowering the cost of maintaining standards. It can flag inconsistencies, suggest improvements and reinforce agreed conventions across large codebases.
This is particularly relevant for modern, data-intensive applications where distributed architectures and complex data models leave little margin for error. AI can support good engineering principles by improving validation, surfacing edge cases and increasing visibility into how systems evolve.
The debate around vibe coding muddies the waters about AI in our industry – the question is not whether AI will be part of software development. It already is. The differentiator will be how seriously organisations treat governance, architecture and long-term maintainability in an era of accelerated output.
By Paul Done, Field CTO at MongoDB
- Data & AI