Vibe Coding!

Some time back, I shared my thoughts on core banking systems—their evolution and where they are likely headed. While that thinking continues to mature, the technology landscape around us has been moving at a pace that makes even recent assumptions feel outdated.

There’s an old cliché: “If you understand something well enough, it’s probably already obsolete.”  That cliché has never felt more accurate.

To put this in perspective, just a couple of months ago, I wrote a brief reflection on vibe coding and the future of software development. Within weeks, it already felt dated—forcing me to discard it and rethink the problem entirely.

At Affinsys, we’ve experienced this first-hand. We built a Low-Code / No-Code (LCNC) platform from scratch—rolling out features quickly, enabling multiple customer journeys, and eventually composing full-fledged fintech applications: onboarding, open banking, CRM, recommendation engines, and more.

But the real inflection point came when we re-armed the platform with copilots - better described as agents.s

These weren’t generic coding assistants. They were context-aware agents, embedded in the platform and aligned with the rest of the platform:

• Natural language–driven analytics and dashboards for business and ops teams
• Building screens and applications via prompts—grounded in domain models, APIs, and core constraints
• Transform APIs (this layer could almost mimic a lightweight service bus)
• Agents that generate campaign content within regulated campaign managers, given the context and cohorts.

At first glance, this might look like vibe coding. But there are subtle (and not-so-subtle, if you look closely) differences. In production-grade systems, the distinction amplifies a lot.

Vibe coding, in my view, is generating code or UI without deeply reviewing the output. It’s something even a non-engineer can do today. Is that bad? Not really.

But here’s the critical question to you: Would you deploy this directly to production?

If the answer is yes, then you’re probably vibe coding. These could be great for your weekend project or a demo. But, not really something you would push directly into your org Git repos.

Now contrast this with assisted development.

Here, developers remain firmly in control. Outputs are reviewed, refined, and iterated. Prompts are enriched with domain context—your frameworks, design patterns, data models, security constraints, and regulatory realities. This is where AI stops being an unchecked author and becomes a collaborator, and starts becoming an engineering force multiplier.

Looking ahead to 2026, I suspect we’ll move beyond vibe coding toward something like “Tribe Coding”.

In this model, entire organisations use shared, connected agentic tools that understand institutional knowledge—core schemas, integration patterns, control frameworks, and compliance boundaries. Prompts won’t be personal hacks anymore; they’ll be versioned assets - reviewed, checked in and governed just like code, 

Glad to hear your thoughts!

AgentOps anyone?

Affinsys AI (BankBuddy.ai, TelcoBuddy.ai, InsureBuddy.ai)

#CoreBanking #FinTech #DigitalBanking #LegacySystems #AIinBanking #OpenBanking #BankingTransformation #Affinsys #AgenticAI #FinancialServices #VibeCoding