Amp and Vibe coding

Software Engineering is a significant domain with various challenges. One such challenge is the shortage of professionals capable of producing the code required to build the systems that companies need. Generative AI has the potential to assist in this area but also introduces new challenges. There are several important considerations regarding the use of AI-generated applications, including how to manage them, how to test these applications, and the appropriateness of using generative AI tools for code generation. Due to the significant market demand, numerous companies have developed applications designed to assist developers with Generative AI, aiming to expedite processes, reduce costs, enhance quality and achieve a faster time-to-market.
In February 2025, “Vibe Coding” was introduced as a new method of software engineering where AI takes the lead. While this approach may be useful for quickly creating MVPs, its effectiveness in enterprise environments remains uncertain and dubious.

The Revolution

Before Vibe Coding gained traction, several key innovations had already laid the groundwork. Before 2022, code development was mostly a manual activity: navigating through the repository
using an IDE, editing files, adding new files and writing code by hand, verifying and testing the application.

This shifted after 2022 with AI assistants. AI assistants support developers with code suggestions, and auto completion. The developer’s workflow has not really changed, but now, during
editing, the integrated AI assistant, autocompletes lines of code and the user hits the “TAB” button to accept the change suggested by the AI assistant (also known as ‘tab to complete’).
Additionally, a chat feature is available that enables developers to pose queries and copy responses directly to the IDE.

In 2024, agents where introduced which changed the workflow a lot, definitely in combination with AI assistants that still have a prominent place in the IDE. Now users navigate through the repository, identifying what needs to be changed, or even asking the agent to help them plan the changes, followed up by requesting the agent to do the work. The agent proceeds to research the best approach for completing the task, edits the changes, and executes this. Meanwhile, a developer can still intervene by making manual edits, or work in combination with an AI assistant (for code completion and suggestions).

Vibe Coding

Vibe Coding is the idea that you no longer write code, you describe it. Think of it as prompt-driven software development. You map out what you want in natural language, prompt a Generative AI model with it, and it generates the code for you. You test it (for example in your browser), “accept all,” and move on. If it fails? You prompt the error. The model fixes it. Rinse and repeat. “And forget that the code even exists”. No coding experience required.
One of the significant challenges with Vibe Coding is the difficulty in understanding its output. While it may appear to function correctly on your screen, this does not guarantee the absence of vulnerabilities within the generated code. If you do not comprehend the code, you cannot effectively verify it. Most models and tools do not include the latest information, meaning
recent security vulnerabilities are not considered unless proper measures are implemented. While this may not pose a substantial issue for small MVPs or POCs, it can present considerable
risks for production-level, enterprise applications.

Amp Coding

This is where Amp Coding comes in, standing in contrast to Vibe Coding. Amp Coding, also known as Amplified Coding, places the expert in control. Instead of “accept all,” it advocates for “review with intent.” Leveraging Generative AI, users can generate code, seek assistance, or request documentation. Ultimately, the responsibility lies with the user, whose name appears on the commit. Therefore, thorough review using Github-like options provided by most tools is essential. Collaboration with one or multiple agents ensures the development of production-ready software. In Amp Coding, it’s the expert who leads, not the agent or model.

The image shows a user collaborating with an agent. While using the IDE, users can access AI assistant features like code suggestions and autocompletion powered by Generative AI. Developers review and approve the agent’s changes to ensure quality standards are met.

So, No More Vibe Coding?

Vibe Coding can be a useful approach, particularly for ideation. Although it may not be suitable for production code, it is effective for creating working prototypes or mapping out ideas with stakeholders, board members, or management. Once the generated code meets expectations, it can serve as a starting point for the product team. Quality Engineers can validate the requirements
more easily, and developers can better understand the requirements. This enables Product Owners or other stakeholders to communicate effectively with developers.
For production-ready code, a higher level of quality is required. Experts need to validate the results or, ideally, an expert should guide the AI in the right direction. Involving experts in the workflow
ensures that quality remains a key element.
In summary, Vibe Coding is beneficial for prototype development and idea mapping but requires expert validation for production-level applications to ensure the highest code quality, which is where Amp Coding comes into play.

Vibe Coding Tools

Vibe Coding tools, also known as “AI Engineering Tools” are primarily designed for individuals without coding experience and often emphasize Vibe Coding.
Examples of these tools include Lovable, Bolt.new, Base44, V0, among others. Similar to ChatGPT, these platforms typically feature an input field where users can enter prompts. Upon execution,
the platform generates a plan (such as a planning.md file) and initiates development within the environment. Some tools provide functionality for reviewing and editing the generated code. However, most users primarily employ a prompt-driven workflow to develop applications using these solutions.

Many of these tools provide features for editing the frontend in a manner similar to traditional content management systems (CMS), as well as for individual components. Users can select objects and modify them through prompts within the tool or edit text directly in the browser in real time.
These platforms have a variety of use cases. They are suitable for MVPs, prototyping, or small websites, but may become more costly and challenging to modify over time. Developers can use them to generate frontend components and integrate these with their IDE tools. Stakeholders without coding experience can also use them to prototype ideas and share those concepts with a development team.