09/07/2026
Our recent Learning Lounge session, "Code with an AI Co-Worker," was led by Ranjith N M and Jomon C Antony from our Web Platforms team. They demonstrated a full loop where an AI agent picked up a ticket, worked through the codebase, wrote and debugged the change, pushed to Git, and raised a merge request.
While getting that to run was easy, the session highlighted that the design decisions around it are what matter.
Reaching tools through MCP, so the setup holds when we change tools.
Reading conventions and architecture notes before writing, since output quality follows from the context it is given.
Running in a manner which keeps the audit trail clear, and reviews sharp.
Runing on codebases only with approval, on endpoints with no training and no retention.
An experienced engineer still owns what ships, including quality, security and performance. The agent prepares the change and the person responsible, reviews thoroughly.
These are patterns we first apply in our own work, so we can validate them in practice before recommending them to clients.
