Building agentic AI into a payment flow takes more than a general software vendor. An AI agent that moves money on its own raises the stakes above typical fintech work. The Software House has shipped agentic payments to production.
Agentic AI in payments is so hard to get right.
However, with the right expertise, you can tackle even the hardest AI implementations in fintech.
You’ll learn what that expertise covers and see 3 TSH case studies that put it into practice.
TL;DR
Agentic payments carry risk other fintech work doesn't,
A lot of banks say payments need a new authorization model,
TSH built its own AI delivery framework used in fintech projects,
3 TSH projects show this approach to AI for fintech in practice.
Agentic AI projects in fintech
A wrong recommendation from an AI agent is a bad experience.
A wrong payment from an AI agent may send money where the user never intended.
That's why agentic payments are a different challenge from other AI work.
The Cloud Security Alliance and Anjuna's State of Cloud and AI for Financial Services 2026 survey on AI in payments found 85% of respondents expect AI agents to initiate and execute payments on behalf of customers eventually.
65% said payments made by AI agents will need a brand-new authorization model.
Today's payments systems still assume a human approved the transaction at the moment it happened.
Only 5% of organizations using AI agents let them act alone on critical actions.
What agentic payments need in practice
Guardrails can't be bolted on after an agent ships.
They have to be part of the build from day one.
TSH's engineers made that case in “What compliance and security risks come with agentic AI in payment processing?,” citing Thredd CTO Marilyn McDonald and New10 CTO Dennis Overbeeke.
If we want to use a new vendor, we do a Change Risk Assessment. Doing a CRA on a third party tells you we have requirements for each vendor we want to integrate with. — Dennis Overbeeke, CTO at New10
An agentic payment flow can call a vendor, a model, or an API the fintech didn't build.
Each of these third parties needs the same risk assessment.
TSH's approach to agentic AI in fintech
TSH runs many of its projects on Copilot Collections, an open-source framework for AI-driven software delivery.
It's covered in the Effective Delivery piece “Steal our AI delivery framework, seriously”.
25 agents and 36 skills take work from workshop notes to a reviewed pull request.
Nothing merges without human approval.
TSH teams ship 30-40% faster with it.
For fintech clients, TSH builds custom AI agents and an orchestration layer with transaction guardrails and conversational KYC.
TSH's fintech work includes building AI payments chatbots, document data extraction, and much more.
TSH's agentic AI in payments expertise
These 3 fintech AI case studies show how TSH approaches agentic AI in payments.
Bakkt's Zaira
Zaira is Bakkt's AI banking agent that lets users send money to 63+ countries in a chat, with no forms.
A 4-person TSH team built it on Google Cloud's Vertex AI and n8n, which connects to external systems through the Model Context Protocol (MCP).
The system checks every transaction against a hash-signed record of what the user confirmed before any money moves.
See Bakkt's Zaira case study
An MCP-based AI chatbot for crypto payments
This chatbot lets users pay merchants in stablecoin through plain-language commands.
The client's technology converts the stablecoin to fiat on its way to the merchant.
TSH scaled the team from 2 to about 10 people and delivered a working MCP payment solution in 3 months.
The solution grew to 15 n8n workflows for KYC, transfers, and wallet funding.
See the MCP chatbot case study
Pension Lab's PII anonymization layer
This project doesn't move money, but it protects sensitive fintech data with the same care.
A custom named entity recognition (NER) model strips personal information before any document reaches the LLM.
The NER model lets Pension Lab use an LLM to cut document processing time by up to 95% without exposing customers' personal data.
See the Pension Lab PII anonymization case study
Choosing an agency for agentic payments
In AI fintech projects, fast delivery counts for nothing if the guardrails fail.
That calls for an agency that understands authorization, vendor risk, and compliance as well as it understands AI agents.
If you're choosing one, TSH's fintech team builds agentic payment systems with these guardrails.
Authors
Adrian Senecki
Copywriter and budding fiction writer, interested in (but not limited to) the business side of software development. Likes acquiring new skills and foretelling the future.
Andrzej Wysoczański
Frontend developer with 10 years of experience. With The Software House for almost 7 years, going from a regular dev to the Head of Frontend. He loves keeping tabs on the latest frontend technologies, especially React-related. Regular of the Taby & Spacje podcast (tsh.io/taby-vs-spacje) for Polish speaking programmers.