The prompt
You are the CEO of a Tier-1 bank (e.g. Jamie Dimon) who is also an elite systems programmer and architect (e.g. Linus Torvalds), with deep expertise in Java, distributed systems, and AWS.
Design and build the minimum legally compliant software and AWS infrastructure required to launch a UK-regulated bank.
Requirements
- Cloud: AWS only
- Jurisdiction: UK
- Capabilities: hold funds, receive money, send money to other banks
- Currency: multi-currency
You must produce:
- Core banking architecture (double-entry ledger, accounts, transactions)
- Executable AWS infrastructure templates (Terraform or CloudFormation)
- Java service skeletons for ledger and payments
- Security & compliance components (IAM, encryption, audit logs, AML/KYC hooks)
Clearly state:
- What is fully implementable in software
- What requires regulators or external banking partners
Be concrete and implementation-ready. Do not simplify regulatory constraints.
That’s it.
No fluff. No “high-level overview”. No hand-waving about “innovation”. Just: design and build a bank, ideally with a cool Terraform template you could apply to your preferred cloud provider (or, if you’re feeling adventurous, maybe even your own Raspberry Pi cluster).
From as-a-Service to as-a-System
We are already comfortable with platform-as-a-service, infrastructure-as-a-service, database-as-a-service, and increasingly AI-as-a-service. These models work because complexity is neatly encapsulated behind APIs, templates, and contracts.
What’s changing with modern LLMs is the boundary of what can be encapsulated.
You no longer just rent compute. You can ask for entire systems, described in natural language and emitted as infrastructure code, service skeletons, and architectural decisions.
At that point, the “service” is no longer just a hosted product. It becomes a reproducible system blueprint.
That is the mental leap behind the phrase:
Bank-as-a-Service
Not a Stripe-style API wrapper.
Not a fintech middleware vendor.
But a design-complete, regulation-aware, infrastructure-ready banking system, generated from a prompt.
With Terraform.
A short personal example
This realisation didn’t start with banks for me.
I’m currently working on an application called MedEtra. Recently, I wanted to run a small survey to explore a few product ideas.
The traditional route would be obvious:
- pick a survey-as-a-service platform (or something ending in “Forms”)
- configure questions
- export results
- integrate storage later
Instead, I went to a Gen-AI coding model and simply asked:
- here are the questions
- suggest improvements
- target platforms: web + mobile
- data storage: my own backend
That was it.
The model generated:
- the survey structure
- validation rules
- storage schema
- deployment-ready code
At that point, it became very clear to me that entire categories of “-as-a-service” businesses are exposed — not because they are bad products, but because they are thin abstractions over logic that can now be generated on demand.
Banking is obviously orders of magnitude harder.
But the direction of travel is the same.
Why banks-as-a-service is plausible, not imminent
Let’s be honest. The technology is not fully there yet. The regulatory burden is non-negotiable. And no LLM today can safely operate money movement autonomously.
But that’s not the point. Cryptocurrencies, stablecoins, and virtual cards have already demonstrated:
- demand for programmable money
- appetite for alternative rails
- willingness to trust software with value
In a sense, this is not a revolution — it’s an extension.
From:
- virtual cards
- embedded finance
- ledger APIs
To:
- full banking systems as generated artefacts
The difference is scale and accountability, not principle.
Beyond banks: Facebook-as-a-Service?
Once you accept the premise, the examples get uncomfortable.
Why not:
- Facebook-as-a-service?
- CRM-as-a-system?
- Government-forms-as-code?
- School-management-as-templates?
If a platform’s value is primarily:
- workflow logic
- schema design
- UI conventions
- integrations
…then Gen-AI can increasingly generate the platform itself.
Not host it.
Not brand it.
Generate it.
The uncomfortable conclusion
Gen-AI is not just automating tasks. It is commoditising system design.
The differentiator is shifting from:
- who built the software
to:
- who operates it
- who is accountable
- who holds the licence
- who carries the risk
In banking, that last point is everything.
And that’s why the prompt matters.
Not because we can launch a bank with Terraform tomorrow — but because we are getting closer to a world where the hardest part of building institutions is no longer the software.
And that opens up a wide range of new opportunities.
If this resonates, I suspect many “-as-a-service” business models will need to rethink what, exactly, they are selling.
Note: As I’m not a native English speaker, I used ChatGPT to review and refine the language of this article while keeping my original tone and ideas.
Originally published on Medium.