A Bank Built Its Own Coding Agent. It Took Off — Felipe Felisola Caso & Francesco Galletta, Monzo
Summary
Monzo details the successful development of 'Agent Chip,' an in-house AI coding agent that has become a core component of their engineering platform. Instead of focusing solely on the LLM technology itself, the key success factor was building a robust internal platform—the 'lots of little multipliers'—which included an MCP gateway, a sandboxed agent orchestration system, and the ability to treat the agent as a composable building block. The agent now authors over 10% of all merged PRs and runs over 1,800 tasks daily, enabling automation in areas like incident response, data quality monitoring, and allowing non-specialists to contribute code drafts.
Key takeaways
-
Focus on Multipliers, Not Just LLMs
The success of the agent was enabled by pre-existing platform capabilities (e.g., alerting, monitoring, staging environments, queuing abstractions) rather than just the LLM itself. These multipliers allowed the agent to function effectively.
-
Composability as a Building Block
By treating the agent as a composable building block, Monzo can connect it to various internal systems (Slack, GitHub, Linear, and over 3,000 services), driving value across product features and business outcomes.
-
Controlled Agentic Development
The architecture uses a sandboxed, isolated namespace for remote agents, ensuring that the non-determinism of the AI is controlled and managed within a regulated banking environment.
Technical details
-
MCP Gateway
351s
The Multi-Context Platform (MCP) gateway was developed to provide easy, controlled access to internal knowledge sources and services. It manages authorization and authentication, allowing agents to connect safely to internal tools.
-
Agent Orchestration and Sandboxing
450s
The agent operates within an isolated namespace and sandbox, which is crucial for a regulated bank. The architecture uses an agent image to define exact tools and access (CLIs, environment variables), and a proxy to manage egress, ensuring controlled access to repositories and code.
-
Agent Use Cases
600s
The agent is used for advanced tasks including: 1) Incident Mode (understanding logs, traces, and connecting to services during an incident); 2) On-Call Data Quality (observing failures and triggering investigations); and 3) Proactive Monitoring (identifying potential data issues from back-end changes before batch execution).
Mentioned resources
Channel & topics
Watch on YouTube · Back to latest
This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.