Why would you need Eximee if an AI agent can generate an application?
Eximee is a collection of decisions made once and recorded in the platform, which code generation would otherwise have to make in every application from scratch.
The agent shortens the path to a working solution, and Eximee takes responsibility for what happens once the solution is running.
An agent can take one of two paths
The agent writes code from scratch
An agent can write the code for an entire application from scratch. In this case, every decision about session, permissions, and integrations has to be settled anew in that application.
The agent builds on the platform
An agent can build a business application on the platform: a process, forms, and rules that run on a ready-made foundation. This is how Consdata teams already build entire solutions.
On the platform, the agent’s task is easier: the platform narrows its field of work to known elements, so the agent doesn’t have to touch complex matters such as session management, permissions, or integration configuration. They are already solved.
What’s underneath
The visible part of an application is a screen, a form, a working process. Underneath lies the rest of the banking solution: integrations, data protection, accessibility, monitoring, versioning, audit trail, change control, maintenance, channels, and session. Someone has to decide how each of these is handled. With agent coding, that decision is made anew in every application, but with Eximee, it is made once, in the platform.
Solution layers
How are roles divided, from the agent’s idea to a working process?
Solution creation
The AI agent prepares changes. Each of them comes with evidence and a human who approves it.
Execution
Eximee, a low-code platform with a process engine, runs the solution and oversees its operation.
Agent in the process
An agent call is an ordinary process step, run within the bank’s infrastructure.
AI in building solutions
The agent works from the project’s recorded materials.
Consdata teams use AI agents when building low-code applications, and a human approves every change. Before showing a result, an agent may cycle through change, verification, and correction many times. We carry the same discipline of evidence and human approval over to the platform.
AI inside the process: process plus AI agents
We don’t build a supervising agent that manages other agents.The process manages the course of the case, and the AI agent is one of its steps, just like data verification or application approval.
As a result, the agent’s behavior has clear boundaries, is repeatable, and can be reproduced during an audit. The bank runs the agent in its own environment, on an AI model it chose itself.
The principles we follow when working with agents
A deployed mechanism has an owner
We consider a mechanism deployed when it has an owner, a permanent place where it is recorded, and a verifiable effect.
Templates drawn from experience
Templates are built from experience that practice has already verified.
Every error tightens the process
When an error appears in how a process runs, we don’t just fix it once. We turn it into a new rule or an additional check so the situation doesn’t happen again.
How it fits together
The agent proposes
The agent prepares a change or proposes a process that is readable to and approved by a human.
Humans approve
The human makes the decision; the agent carries it out.
Eximee executes and controls
The approved process operates on the process engine, which lets you run, observe, and justify it.
Fit within the bank’s applications
Integrations
Eximee solutions integrate with the bank’s existing applications: online and mobile banking, as well as internal systems such as CRM or the branch application. The integration with those applications is built and tested once, and subsequent solutions reuse it.
Interface
The platform enforces interface and behavior patterns centrally rather than within individual projects.
Channels
The platform provides omnichannel capabilities: a case started in one channel can be finished in another, and an employee can take it over from the customer or from another employee.
Maintenance over the years
When applications are generated separately, changing a rule, for example one concerning accessibility or data protection, means reviewing each of them. On the other hand, in the platform, the change is made in one place.
Shared mechanisms
The process engine provides shared mechanisms: incident handling and retries from the console, work that survives the failure of a single server, process history, and process versioning independent of service code.
Freedom of choice
The bank can return to manual or mixed work at any time, and the runtime rules apply to every change, regardless of who made it.
Questions to ask before letting an agent into a process
Five questions worth answering before an agent enters a process:
Frequently asked questions
How does an agent enter a process?
As an ordinary process step, with the same constraints as any other call.
Do you need another agent to supervise agents?
No. The process directs the agents’ work.
Who makes decisions: the human or the agent?
The owner makes the decision; the agent carries it out.
Where does the agentic step run?
Within the bank’s infrastructure, on a model chosen by the bank.
How does an AI service differ from an AI agent (agentic AI)?
An AI service is a single call with a contract defined up front. An agent chooses its own actions and tools, and the number of calls is not known in advance.
What do human in the loop and human on the loop mean?
These are two oversight patterns. In the former, a human approves a specific output; in the latter, the human supervises the agent’s operation and steps in when needed.
