I’ve built financial models for over a decade, mostly with various specialized Financial Planning & Analysis (FP&A) products for customers cross manufacturing, retail, utilities, and financial services. When I first heard about “vibe coding” in this context, my initial reaction was skepticism. How on earth should prompting AI generate proper applications from scratch? But after seeing the rapid development of AI capabilities, I realized it’s a shift that could reshape how we think about building and buying FP&A solutions, even if it’s far from certain where we’ll end up.
How does natural language-to-code work in practice?
Vibe coding means describing what you want in natural language and letting AI generate the code. No traditional programming required. You iterate by chatting with the AI: “make the dashboard blue,” “add a revenue forecast module,” “read data from this Excel file and load results to that.” The AI writes the code, you test it, and you refine it through conversation.
Code generation is one area where current GenAI tools are often strong. It can produce working applications faster than any other medium. For finance professionals who understand planning logic but never learned to code, this suddenly makes custom tool development accessible. But the approach is not free from threats.
Vibe coding in FP&A works well for simple requirements but breaks down when it gets complicated
Vibe coding is currently hyped, and finance teams are starting to experiment. But how do you engage with it smartly? Simple requirements go fast. A basic budget input form, a departmental dashboard, or a data import script can materialize in an afternoon or even in a few minutes. Complex financial logic like multi-year forecasts with loans, capex, depreciation, and consolidation rules, that’s a completely different story. The more freedom you give the AI, the more thoroughly you need to verify what it produces.
The hype creates pressure to move quickly. But speed without understanding creates technical debt. Think of it as the ‘grinchy’ sibling of vibe coding: the mess that accumulates when speed outpaces understanding.
Security, integrity, and maintenance – the hidden risks that come with speed
AI has no judgment. It will take dangerous shortcuts and won’t prioritize security or data privacy by default. You cannot sue it, so why would it choose the harder path? Performance, security, and calculation integrity don’t happen automatically. They all require deliberate design and verification.
Maintenance is the hidden cost. Code generated in minutes still needs support, updates, and governance. You can produce code faster than you can review it. If business users generate applications with the best intent but without technical oversight, you’ll accumulate unmaintained tools that nobody fully understands. When something breaks in production, who fixes it? And we know that story. Many applications were built Excel using extensive VBA. When they ceased to work properly, business users were quick to delegate fixing, refactoring, and maintenance to IT. And we see it happening with Power BI too. Do we really need to do that again?
Technical debt is the top concern. Fast code isn’t automatically good code. It needs architecture, documentation, and someone who understands it well enough to modify it six months later. We risk creating disposable code like fast-fashion for software. The Primark-ization of applications. That might work for throwaway prototypes, but not for solutions that power your most critical financial close processes.
How can teams use guardrails to build quickly without creating a mess?
You need someone who can understand and verify what the AI produces. Business users can use vibe coding for small, contained changes like adjusting a formula or tweaking a calculation that doesn’t cascade into other systems. But building new platform capabilities or complex features requires someone who understands application development and can provide technical oversight. Establish technical review before anything touches production data or drives critical decisions.
But is reviewing really quicker than building “properly” from scratch?
Frameworks that provide stable foundations are valuable. If the basic infrastructure providing security, data access, calculation engine, is proven and maintained, then building on top becomes safer. Some FP&A products now include agents that generate planning logic within guardrails. That’s vibe coding with a lifeline, and it might be the right balance as it brings flexibility and speed without creating huge risks. We’ll be watching this space closely.
Data management with its imports, transformations, and integrations, consumes most of the time in FP&A implementations. This is where vibe coding shows high potential. You’re often working within established environments with known data structures. Generating ETL scripts or data validation rules is less risky than generating core planning logic.
Test quickly to understand. You don’t learn to swim by reading about it. Build a prototype, break it, understand what went wrong, and iterate. But treat it as a learning exercise, not production deployment. It may serve as a valuable blueprint and help you to sharpen requirements as you go.
Don’t repeat the mistakes we’ve made before. We’ve seen technology shifts from spreadsheet chaos to premature cloud adoption. We’ve oscillated between “this changes everything” and “we missed the boat”. Vibe coding is powerful, but it’s not a replacement for understanding your requirements, your data, and your processes.
How will AI-built software change vendor landscape, pricing, and consulting needs?
New vendors will emerge who build their products using AI-generated code. This lowers barriers to entry. Expect more competition in simple data collection and reporting tools as these are easiest to generate and will proliferate. This means there will be a wider range of software to select from, but the quality of new offerings will require thorough vetting.
Price pressure is coming. Good for you, maybe not so for the vendors. When the make-or-buy decision shifts because internal teams can generate applications faster, vendors will need to justify their value differently. Consulting services will shift. Simple configuration and customization work becomes less valuable when customers can do it themselves. The flexibility gained will offer huge benefits to them. On the other hand, strategic work such as understanding complex requirements, designing scalable architectures, and integrating across systems, becomes even more valuable.
It’s too early to predict where the market is going but we will feel an impact soon
This topic came up in an intense internal discussion with my colleagues, and I’m still processing the implications. Having implemented planning models across many industries and products, I keep asking myself: where does this genuinely add value, and where does it create risk we’re not prepared to manage?
And it raises further questions worth exploring: Who can it do what for? Where do FP&A experts benefit? What changes for ETL developers? Will test-based development be the new default (and what actually is that)?
This post is less a prediction and more a collection of thoughts for finance leaders evaluating their options. Vibe coding will impact how you build, buy, and maintain FP&A software in the future. The question is whether you’ll engage with it deliberately or react to it later.
