Bank of Georgia has moved beyond AI coding assistants to create an autonomous AI developer capable of taking a Jira task from analysis through to production-ready code. Combining multiple AI models with its internally developed Code2Doc technology, AI Employee autonomously authors 40% of the Java codebase for the bank’s digital products.
For software developers, GenAI has become a familiar assistant suggesting code, answering questions, or completing tasks when prompted. Bank of Georgia wanted to go further.
Its AI Employee is a coordinated group of autonomous AI agents designed to operate as an active member of development teams. Assigned a task through Jira, the system independently gathers context, plans, writes, and tests code, and submits a merge request, with a human developer retaining responsibility for final review.
From Copilot to Autonomous Developer
The fundamental difference between AI Employee and conventional coding assistants is autonomy. Rather than requiring a developer to initiate each stage, one Jira assignment triggers an end-to-end workflow.
Four specialised components work together:
- AI-Communicator connects the system with tools including Jira and GitLab.
- AI-Employee coordinates the overall workflow.
- AI-Analyst interprets requirements, gathers context, and creates the technical plan.
- AI-Developer generates the code and unit tests.
The architecture avoids relying on a single foundation model. Google Gemini is used for requirements analysis and technical planning, while Anthropic Claude handles code generation and unit testing, allowing Bank of Georgia to select different models for the tasks to which they are best suited.
Crucially, the system can recognise when it lacks sufficient information. It asks questions directly through the relevant Jira ticket and incorporates the developer’s response before continuing.
“One of the most mature examples of agentic productivity in production globally – truly novel, production-proven, and ahead of the market.”
“A jaw-dropping, highly disruptive production milestone that sets a new global benchmark for engineering automation.” – Judges’ comments
Giving AI the Context to Act
Autonomy depends on understanding the organisation in which the AI operates. Bank of Georgia therefore designed AI Employee to proactively retrieve both business and technical context.
Business information is gathered from Confluence through Rovo, while technical understanding comes from a knowledge base generated by Code2Doc, an internally developed model that automatically creates detailed documentation directly from source code.
This allows AI Employee to operate across different codebases and IT divisions rather than being restricted to a single team or technology. Key capabilities and outcomes include:
- AI Employee autonomously authors 40% of the Java codebase for Bank of Georgia’s digital products.
- Tasks progress from Jira assignment to merge request without human initiation at every development stage.
- AI autonomously performs analysis, context gathering, code generation, and unit testing.
- Multiple AI models are orchestrated according to their specialist strengths.
- Code2Doc automatically turns source code into a technical knowledge base for AI Employee.
- A central dashboard provides real-time visibility and auditability across every stage of each task.
- AI operational costs are tracked at individual task level.
- Human developers retain final review and approval before generated work is accepted.
Building AI into the Team
Bank of Georgia designed the system as a modular platform capable of evolving alongside its technology estate. New capabilities can be introduced independently, while different IT teams can onboard AI Employee and assign tasks relevant to their own domains.
That scalability supports the broader objective of increasing development capacity without requiring a proportional increase in human resources. Routine development work can be delegated to AI, allowing senior developers to concentrate on architecture, complex business logic, and problems requiring creativity and judgement.
The innovation is as much about the operating model as the underlying technology. Bank of Georgia has changed the point at which humans interact with AI, from continually prompting an assistant to assigning work to an autonomous team member and reviewing what it produces.
With AI Employee already responsible for a substantial share of its Java development, the bank is demonstrating what the next phase of enterprise AI could look like by redesigning how human and AI capabilities combine to deliver work.
