Leveraging Artificial Intelligence for Development Optimization
The integration of artificial intelligence (AI) into the Product Development Life Cycle (PDLC) is becoming a pivotal trend for major financial institutions. Alfa-Bank, a leading player in Russian fintech, is actively restructuring its operational processes by deploying AI agents to enhance efficiency and reduce costs. Mikhail Chernov, Senior IT Leader for IT Collection Development at Alfa-Bank, shared insights into their transformative journey.
“Estima”: An AI Agent for Task Estimation
One of the earliest and most successful implementations was the creation of an AI agent named “Estima.” This agent took over the routine yet highly resource-intensive task of estimating development task duration and complexity. Previously, this required weekly 1.5-hour meetings involving 20-25 specialists. These sessions, based on task descriptions, historical data from similar projects, and past evaluations, consumed significant time from developers and tech leads.
After a 1.5-month pilot, “Estima” achieved a 95% convergence with manual estimates. This success allowed the bank to completely eliminate the weekly meetings, delegating estimate validation to tech leads. As a result, Alfa-Bank reported saving approximately 20 person-days per month, significantly freeing up highly skilled professionals.
Expanding AI Agent Applications Across PDLC
“Estima’s” success convinced the team of AI’s practical value, paving the way for other AI agents to be integrated into various stages of the PDLC. These agents are now actively utilized for:
- Working with the Jira project management system
- Automated code reviews
- Optimizing development processes
- Enhancing testing efficiency
- Improving analytical tasks
- Strengthening application security (AppSec)
The comprehensive application of these AI agents enables Alfa-Bank to progressively transform and modernize its entire production process, making it more agile, faster, and more cost-effective. This opens new opportunities for further digitalization and innovation within the fintech sector.
While the reported efficiency gains from AI in PDLC, especially with “Estima,” sound impressive, I can’t help but wonder about the hidden costs and potential over-reliance. Achieving 95% convergence is great, but what about the 5% that might be critically misjudged? How are those edge cases handled, and what mechanisms are in place to prevent a cascade of errors if the AI misinterprets complex, novel tasks? There’s also the significant investment in developing and maintaining these AI agents, which isn’t always immediately obvious in cost-saving narratives.