The TVT Blog | Business Central Specialists

What Is Vibe Coding? Opportunities & Risks for Business | TVT

Written by Alex Williams | Nov 16, 2026, 11:54:53 AM

 

AI is making it possible for more people to build simple apps, internal tools and new ways of working with business data. You can describe what you want in everyday language and ask AI to start building it, without necessarily knowing the programming language or understanding every line of code.

This approach has become known as vibe coding. At TVT, we have experimented with everything from standalone applications to AI-generated interfaces connected to Microsoft Dynamics 365 Business Central and even a Business Central extension. The most important question is not whether AI can write code. It is what happens when the result becomes something your business relies on.

What is vibe coding in simple terms?

Vibe coding is a way of creating software by describing what you want to an AI coding tool in everyday language. The AI generates much of the code, allowing you to build and refine applications through prompts. For businesses, it can be particularly useful for prototyping and exploring new ideas, but AI-generated code still needs appropriate review before it is relied on for important business processes.

What could you build with vibe coding?

Here at TVT, we have used AI to create tools such as a Business Central licence calculator and explored more involved visual applications that interact with Business Central. One of the real opportunities of vibe coding is the speed with which you can move from ‘Wouldn’t it be useful if…?’ to something on screen that people can test.

That matters when you have a smaller problem that doesn’t justify a formal development project: an awkward spreadsheet, a repetitive process or information that would be easier to use if presented differently. Instead of beginning with a full specification, you can start by seeing what the idea looks like.

Prototyping could be the biggest opportunity

Requirements can take a long time to define, and people often discover too late that they imagined different solutions. AI offers another approach: build something quickly and ask whether it is roughly what you are trying to achieve.

The prototype does not have to become the finished application. Its purpose may be to test the concept, clarify what is possible and support a better decision about whether the idea deserves proper development. AI does not simply make software easier to build; it makes ideas easier to explore through software.

When should you be more cautious?

The risk changes when an experiment becomes useful. A small internal tool may gradually become part of an everyday process. What happens if it stops working? Who understands it well enough to fix it?

Every AI-assisted project needs an escape hatch - an alternative solution if the AI approach fails. If you can stop using the tool and return to the old process, an experimental approach may be reasonable. If its failure would prevent you from processing orders, shipping products or completing another critical task, you need a better understanding of how it was built, secured, maintained and supported.

Working code isn't the same as good software

Code can run without errors and still be unsuitable for day-to-day business use. You also need to consider its architecture, performance, security and whether it duplicates something you already have.

Security problems are especially difficult because they may be invisible. A broken button is obvious; poor authentication or inappropriate data storage may not be. As you move closer to operational or commercially sensitive systems, ‘Does it work?’ cannot be the only test.

Can vibe coding work with Business Central?

We experimented with a visual shipment-planning application that showed unshipped sales orders while keeping Business Central underneath. Rather than creating another database, the application used Business Central APIs, leaving the data where it belonged.

Business Central also continued to govern the business logic. Authentication used Microsoft Entra ID and delegated permissions, so users worked through their existing identities rather than a separate login. This suggests a useful model: use AI to explore new experiences around Business Central while keeping trusted data, permissions and rules at the centre.

Can AI build a Business Central extension?

We also asked AI to create an extension that applied a discount to VIP customers. It generated the AL objects, logic and permissions, and the extension compiled and published to a sandbox. Testing nevertheless exposed a bug, showing why compilable code is not automatically production-ready code.

There was also a more important question: did the extension need to exist? Business Central already includes established pricing functionality. An experienced consultant or developer might have challenged the requirement before writing any code. If standard functionality already meets the need, configuration may be safer than custom code that must be tested, upgraded and supported. AI can answer the question you give it; human experience still matters when deciding whether it is the right question.

AI can expand your capacity or your ability

For an experienced developer in the AL programming language, AI may expand capacity by producing code more quickly. The developer can still review the result, recognise a poor approach and investigate failures.

For someone who does not know AL, AI expands ability by enabling work they could not previously do. That is exciting, but it makes independent review more important because the user cannot assess the output in the same way. The more important the outcome, the more this distinction matters.

Does AI replace developers?

Our experiments suggest that the developer’s role will change rather than disappear. If AI handles more code generation, human expertise becomes even more important in deciding what should be built, choosing the architecture, protecting data, assessing security and recognising when existing Business Central functionality is the better option.

At the same time, business users can bring a tangible prototype to the conversation rather than only a description. The discussion can then focus on whether the prototype solves the problem in the right way and what would need to happen to make it dependable.

How can you use vibe coding without unnecessary risk?

First things first, make sure you experiment with small problems that have never justified a development project. Build a prototype for a difficult-to-explain idea or a more visual Business Central process. But make sure you keep asking where the escape hatch is.

As the application becomes more useful, consider whether it is also becoming more important. Check where its data lives, how users authenticate, which systems it connects to and what would happen if it disappeared tomorrow. Before relying on it in production or for a critical process, arrange an independent review of its code, architecture, security, testing and support needs.

Where does TVT fit?

We want our customers to explore what AI can make possible. The important point is recognising when a promising experiment needs a structured route towards something the business can safely rely on.

If an experiment has sparked an idea for improving a Business Central process, our AI Proof of Concept service provides a structured way to test it before committing to a full implementation. For a fixed price of £2,500, we define the process and success criteria with you, build a working prototype in your Business Central environment, and test and refine it with your feedback. You then receive an implementation proposal with estimated costs and timescales.

The proof of concept establishes whether the idea is viable and valuable. It does not deliver a finished application. If you want to proceed to a finished application, a separate Fit and Finish phase will address production-standard code, security, integration, testing, documentation and deployment. There is no obligation to continue.

Ready to explore your AI idea?

If you have a Business Central process you want to improve, contact us to find out more about our AI Proof of Concept service.

Frequently asked questions about vibe coding

What is vibe coding?

Vibe coding is an approach where you describe what you want to an AI tool and allow it to generate much of the underlying code. You continue developing the application through natural-language prompts rather than working directly with the programming language.

Is vibe coding suitable for business use?

It can be, particularly for prototypes, experiments and small tools with an easy fallback. As an application becomes more important, its architecture, security, testing, maintenance and support need greater scrutiny.

Can you use vibe coding with Microsoft Dynamics 365 Business Central?

Yes. AI-assisted applications can work through Business Central APIs while data, permissions and business logic remain within Business Central.

Can AI create Business Central extensions?

Yes. AI tools can generate AL code, but you should first confirm that custom development is necessary and independently review anything intended for production.

Is AI-generated code secure?

Do not assume so simply because it works. AI-generated code may contain vulnerabilities or poor architectural choices, so important applications require appropriate testing and independent review.

Will AI replace software developers?

It is more likely to change their role. AI can generate more code, while human expertise remains essential for understanding requirements, choosing architecture, assessing security and taking responsibility for the finished solution.

What's the difference between vibe coding and AI-assisted development?

Vibe coding usually means guiding development mainly through natural-language prompts, sometimes with little involvement in the underlying code. AI-assisted development is broader and includes developers using AI to accelerate work they already understand.