Software at the speed of chat

The term “vibe coding” was coined by AI researcher Andrej Karpathy, as AI acts as an active collaborator. Advanced versions use AI agents to collaborate more deeply—writing and running self-tests, or executing multi-file editing across the codebase, enabling rapid prototyping and quick deployment.

The popular tools for vibe coding include Cursor (for professional developers who want an integrated development environment that combines various tools into a single graphical user interface), Replit Agent for beginners, Windsurf/Cline for agentic support, Google AI Studio which leverages Gemini, and the popular Claude Code from Anthropic.

Educational and other imperatives

Considering the developments in making AI tools easier to use, one could argue that structured learning isn’t essential. What’s clearer is that mastery over all tools is a long shot, considering their abundance, and because that type of proficiency is transient. An education grounded in enduring principles is more durable. Like conceptual data flows, precision in semantics, skills to audit code, and enduring clarity of thought.

With AI handling all the grunt work, such a curriculum also signals the shift—from needing more entry level coders to more experienced orchestrators.

Investors are beginning to value product visioning, the scale and depth of the problem being solved, and market timing, more than before.

An aside: It is advisable to be disciplined about vibe coding work hours, as developers transitioning from traditional coding to vibe coding reportedly experience a high—much like having acquired a superpower of sorts—and are inclined to work much longer hours, which could lead to burnout. 

Upsides and downsides

To reiterate, entry barriers are collapsing, allowing non-technical founders to rapidly build functional software solutions at a fraction of the cost, signalling a shift away from the usual, jejune URL path. Simultaneously, as developers become orchestrators, the focus is shifting to system design and user experience. Lean tech teams at nascent startups are now building products overnight, changing the economic equation substantially.

While an AI agent creates code almost instantly, it usually lacks a global understanding of the system. Redundant libraries could be added. Over time, a vibe-coded solution could well become a black box where neither humans nor machine understand the interconnected dependencies. When AI’s context window can no longer decipher the code created, it could lead to a “vibe collapse”. Sometimes, without a complete grasp of the business context, or because of prioritising getting the app to work, oversight pertaining to compliance or original security assumptions could also occur. Expert guidance could help preclude such hurdles.

Maintainability could be an issue too when a complex bug surfaces, owing to the cognitive ownership not residing in human minds.

Governance and cybersecurity risks

Studies reveal that AI-generated code has some inherent cyber vulnerabilities, arising from vibe coders sacrificing safe levels of threat modelling and authentication for speed. This is accentuated when the audit logic is weak, and apps deployed look perfect, but invite AI-driven exploits by hackers. Attackers are now getting smarter, concealing commands inside requests, using true or false questions to reconstruct the database structure, or making the database send information to servers they control for stealthy exfiltration of sensitive credentials.

The natural consequence of coding at the speed of chat is that human oversight becomes the bottleneck, unable to catch all security flaws or compliance drifts across vast codebases. Agentic governance could be the answer, with specialised AI supervisors playing the role of real-time gatekeepers. Vibe coding without digital guardrails is an invitation to trouble.

Scaling paradox

Coders say that building the first 80% of a solution or app, using vibe coding, is incredibly easy, and building the rest, involving scaling, edge cases, and integration, equally complex. That’s also why scaling beyond the initial vision could prove to be a limitation. Another aspect to be wary about is that while AI is great at local optimisation, it may take shortcuts that contradict architectural decisions taken several prompts before. This recency bias causes drift, creating a patchwork of conflicting logic over time. And when a bug is encountered, the tendency of AI to patch the symptom rather than refactor the root cause compounds later navigation. One way of tackling this could be through modular orchestration, breaking the app into isolated, modular components and then integrating them, a more elaborate path. Scaling certainly separates the hobbyist from production grade coding—which demands experienced, patient, nurturing hands and minds.

Creating value

Vibe coding is a major structural realignment vis-à-vis creating value from software, a tectonic shift from syntax-centric programming to AI-assisted software development that prioritises human thinking. And a profound shift away from dealing with semicolon placement, function signatures, and compiler errors, as in the past. When human creativity embraces heavy-duty agentic sweat work, it needs hi-fidelity orchestration—very different from traditional, lengthy IT projects with inherent scope creeps—to sectoral specialists aiming to solve real problems more directly with digital interventions.

With the right levels of human scrutiny based on risk profiles and application criticality, appropriate engineering leadership, and security guardrails, the underlying competitive advantage awaits capture.

It’s not about whether to embrace vibe coding, but how!

Disclaimer: The views expressed are the author’s own and do not reflect the official policy or position of Financial Express.