Stitch adds AI agents to its banking platform, with bank-set limits on actions
The company says staff can request changes in plain language and deploy agents for onboarding, credit decisions and collections, while banks control permissions, deployment and model choice.
Stitch AI gives bank staff and purpose-built agents a way to change products and workflows through the services already running Stitch’s core banking system, rather than through a separate information layer. Launched November 18, 2026, the offering makes an agent’s authority configurable by each bank: actions inside its defined limits can proceed without case-by-case approval, while actions above them require approval. Banks can also choose deployment location and model, and switch models without rebuilding the system.
Stitch says sensitive fields—including account and identification numbers, phone numbers and addresses—are masked before a model sees them.
03
Each agent has its own identity, narrow permissions and hard limits, with maker-checker approvals and logs covering questions, actions, approvals and reasoning.
Bank staff can ask AI to do more than retrieve information under Stitch’s new offering: they can request changes to products and workflows. Stitch launched Stitch AI on November 18, 2026, putting a staff assistant, ready-made agents and agent-building tools inside its banking operating system, with actions subject to bank-defined controls.
The technical distinction, according to Stitch, is where those agents operate. They act on the core banking system through the same services that already run the bank’s products. The company says this avoids copying records into another system or maintaining separate connectors. The proposed route from a request to an action therefore runs through the banking platform itself, rather than a separate information tool.
Three routes into banking workflows
Stitch packages that access in three forms. One takes instructions from employees, another supplies agents for named banking jobs, and the third lets institutions build their own. The company describes each as operating within the permissions and approval controls set by the bank.
Copilot is the staff-facing assistant. Employees can use plain language to fetch account and customer data, execute a task or change a configuration. Stitch says it can set up products including accounts, cards and loans, configure limits, pricing, fees and eligibility, and update workflows.
Stitch Agents are ready-made tools for operations including customer onboarding, credit decisioning and collections. The company says they work within existing workflows, permissions and rules. These are packaged around particular banking functions, rather than requiring a bank to construct every agent through the development framework.
Agent Harness is the framework for banks that want to build their own agents. Stitch says institutions can configure, test and deploy them with permissions, approvals, guardrails and audit trails built in. It provides a custom-building route alongside the staff assistant and ready-made operational agents.
Permissions, masked data and a record of actions
Each agent has its own identity, narrowly defined permissions and hard limits, according to Stitch. The controls also include maker-checker approval—a step that separates initiating an action from checking it—and an audit trail. Those controls attach to the agent operating in the system, not just to the employee’s initial request.
Stitch also says it masks sensitive information before any model sees it, including identification numbers, account numbers, phone numbers and addresses. Every question, action and approval is logged, together with the reasoning behind it. The stated purpose is to give risk and compliance teams a record of what happened and who approved it.
A route beyond bank AI pilots
CEO Mohamed Oueida framed the launch around two obstacles: banking systems that cannot accommodate AI projects, and approval requirements that prevent agents from acting without an accountable owner. He said those problems keep projects stuck at the proof-of-concept stage. Stitch’s pitch is to address the system connection and the authority to act together.
Banks also choose where Stitch AI runs. The company says it supports in-country deployment or a private cloud, with the institution selecting the model. That leaves deployment location and model selection as bank decisions within the offering, alongside the operational permissions.
The model choice is not fixed at deployment: Stitch says banks can switch models without rebuilding the system.
Sources
prnewswire.comStitch Launches Stitch AI, Bringing Agentic AI Directly Into Banking Operations
Reader comments
Newest comments first. Replies stay oldest first.