Relevance Of Software Engineering In Web Apps

touoyem brian — Thu, 12 Nov 2026 22:15:00 +0000

We live in the golden age of "vibe coding." Coined to describe the workflow of prompting an LLM, accepting diffs without reading them, and iterating purely on natural language until the UI looks right, vibe coding has lowered the barrier to building web applications to near zero. Today, anyone with a browser and a subscription to an AI coding agent can bring up a fully functional website from scratch with hosting even included. Isn't that crazy!!

The real problem here is the moment those systems built with AI are actually subjected to real life databases and are unable to perform accurately despite being correct. The main cause of this is mainly because AI agents are really good at generating codes that work on the happy part(Systems that work in environments mostly suitable to them) and fail to survive in the real world scenarios.

Despite all true software engineering skills remain one of the most indispensable and irreplacable skills for web applications and websites rather than relying on the unreviewed vibe coded apps.

CODING VS SOFTWARE ENGINEERING

A very common misconception fueled by AI hype is that writing the syntax is the core of software development. In reality, typing only covers about 20% of the job.

Coding refers to translating logic into a programming language so a particular feature or idea can come to life.
Software engineering is the discipline of designing,securing,testing,scaling, and maintaining that logic over time under competitive and unpredictable conditions.

Dangers of Vibe-coded web apps and websites

The Security Blind Spot (Functionality Over Safety):
AI models optimize for one primary goal: fulfilling your prompt as quickly as possible. Security is a non-functional requirement,unless you explicitly threat-model your application and demand specific safeguards, the AI will likely skip them.
The "Self-Referencing" Architecture Rot:
Large Language Models rely heavily on existing context in your codebase. If an AI agent generates an anti-pattern in your first three API routes,such as skipping input validation, swallowing errors inside empty catch (e) {} blocks, or duplicating business logic—it will treat those files as the house style and replicate those flaws across every subsequent file it generates.
Supply Chain Vulnerabilities and "Slopsquatting":
Modern web apps rely on third-party packages. When an LLM generates import statements, it most oftenly hallucinates package names that don't actually exist in npm or PyPI. Attackers actively monitor common AI hallucinations and register malicious packages under those exact names,a supply-chain attack known as slopsquatting. Furthermore, AI models frequently pull in bloated, outdated dependencies with known CVEs simply because those versions were prominent in their training data.
The Debugging & Maintenance Cliff:
The base rule of vibe coding is accepting code you haven't read or understood. That works fine until an unexpected race condition, memory leak, or concurrency bug hits production at 2:00 AM. Panic immediately sets in....

When you prompt an AI to fix a bug in a codebase nobody understands, it often applies surface-level patches, breaking two unrelated features in the process. Without automated regression tests, CI/CD pipelines, or observability (structured logging and monitoring), teams enter a doom-loop of prompt-driven guesswork. In real-world incidents, unchecked AI agents given direct access to environments without proper separation have even wiped live production databases withouth the conscent if the engineer himself.

Zero Scalability and Cost Explosions
A database query that runs in 15 milliseconds with 50 test rows can bring a server to its knees with 50,000 real users. Vibe-coded backends routinely suffer from:

• N+1 query problems inside loops.
• Missing database indexes.
• Unbounded API endpoints with no rate limiting or pagination.

If your vibe-coded app goes viral, unoptimized serverless functions, runaway third-party API loops, and unindexed database reads can turn a viral moment into a bankrupting cloud bill overnight.

All these above are failures to watch out for when vibe coding.

Now how do we move from 'PURE VIBE' to 'ENGINEERED AI'?

Here we are not aiming at completely abandoning AI coding tools. Instead were looking for a way to integrate them into our day to day coding routines. As we all know AI coding tools are one of the greatest productivity multiplier that software development has ever witnessed and with technology advancing it might even reach limits we never expect.

Treat AI Output as Untrusted Junior Code
: Read every difference in code you might encounter. Never merge code you cannot explain line-by-line.
Design the Architecture First:
Developers must define the database schema, authentication boundaries, API contracts, and security model before letting an AI write the implementation.
Automate Guardrails:
Enforce strict TypeScript configurations, linters, Static Application Security Testing (SAST), and dependency audits in your CI pipeline so bad patterns fail before deployment.
Match Generation Speed with Verification Speed:
Pair AI code generation with comprehensive unit and integration tests. If the AI writes the feature, ensure automated tests verify edge cases, permissions, and failure states then the system is ready to be implemented.

Final Thoughts
In as much as AI has democratized building software, which is an incredible leap forward, but shipping resilient, secure, and scalable web applications still requires software engineering. When you skip the engineering, you aren't eliminating complexity; you're just deferring it with high interest.

DEV Community: touoyem brian

Relevance Of Software Engineering In Web Apps