Skip to content
Muthu Bommiah

Initiative · Wells Fargo · 2024 – present

AI-Augmented Software Development Lifecycle

Coding agents brought into a large bank's software lifecycle behind one governed platform: curated context, maker/checker controls and a single gateway for every model call.

Problem

Developers wanted AI assistants; the bank needed the productivity without losing control. Every line of AI-written code still had to meet the same quality, security and audit standards as code written by people.

Approach

  • Built a multi-vendor agent ecosystem rather than betting on one tool, so the bank can swap agents as the market moves.
  • Wrapped the agents in bank-built capabilities: context assembly, a harness layer, search and retrieval, and one gateway for governed access to models.
  • Started in the developer's inner loop, then extended to the outer loop: CI/CD, platform operations, AI-augmented observability and service management.
Flow from spec or intent to a coding agent, then a highlighted verification harness of tests, policy checks and security scans, then human review and production. Failures loop back to the coding agent to fix and retry.Spec /intentCodingagentVerificationharnesstestspolicy checkssecurity scanHumanreviewProductionfails: fix and retry
In an AI-native lifecycle the bottleneck moves from writing code to verifying it. The harness decides what reaches a human, and what reaches production.

Governance

  • Maker/checker controls: agents produce, people and automated checks verify before anything ships.
  • One gateway for every model call, so access, logging and policy are consistent.
  • Automated vulnerability and patch remediation with continuously generated software bills of materials.
  • Pilots proved value on a shared dashboard and passed stage gates before investment scaled.

Outcome

  • A measurable productivity gain across the developer population, tracked on a shared dashboard and corroborated by developer surveys.
  • A governed path that lets new agents and use cases join without building a new control framework each time.

Lesson

Model is a tool; harness is the strategy. Buy the agents, build the governance layer around them.