When Brooks joined IRALogix in 2024 as Senior Director, Head of Software Engineering, the company was a Series C fintech offering IRA-as-a-Service to a growing list of platform clients. The platform had $900M in AUM. The engineering organization was running a traditional SDLC. Both numbers needed to change.
The mandate was uncommon for a fintech of that stage. The board needed three things at once: a platform that could absorb a 10x increase in business-development pipeline, SOC2 Type 2 certification to unlock the enterprise segment that gated further growth, and an engineering operating model that did not depend on hiring at a pace the labor market could not support.
Solving any one of those was a six-month project. Solving all three together, in twelve months, required rethinking how software got built.
The starting position
The engineering team was running the model most Series C companies run: capable engineers writing code against tickets, a code review process that took whatever shape each tribe chose, an architecture board that approved decisions after the fact, and a hiring pipeline that competed with every other Series C in the same talent market.
That model is the right model for the previous era. It is not the right model for a company that needs to ship faster, satisfy a Type 2 auditor, and absorb a new development paradigm at the same time.
The transformation
The work happened in three concurrent tracks.
Track 1: AI-SDLC as the new development practice
The engineering organization migrated from traditional development to a spec-based agentic workflow. The unit of work became the specification. AI agents generated, reviewed, and iterated on implementations against those specifications. Human review moved to the gates where it produced the most leverage: spec approval, architecture review, pre-production sign-off.
The result was a 60% reduction in cycle times and a 40% reduction in defect escape rates. Equally important, every change in the codebase now had a written specification, a written review, and a documented sign-off, which is exactly the audit trail that Type 2 certification requires.
Track 2: MCP and context infrastructure
A small founding design team authored standardized agent context files for every application in the codebase. These files described the architectural patterns the company used, the data models that existed, the security boundaries that could not be crossed, and the conventions that had been established.
The work then expanded firm-wide in partnership with Enterprise Architecture and Platform Operations to meet shared security, observability, and operational standards. The architecture board itself was transformed from a traditional review committee into an AI governance and spec stewardship body, jointly defining agentic coding standards, prompt governance, and model usage policies across all product teams.
This is the precondition that most companies skip. Without it, agentic tools produce plausible mistakes at scale. With it, the productivity gains compound.
Track 3: Organizational redesign
The engineering organization was restructured into capability-driven product teams, anchored by a north-star architecture: API-first, custodially agnostic. The architecture choice meant the platform could serve any custodian a client wanted to use, which removed a frequent objection in sales conversations and accelerated client acquisition directly.
A transformative digital account opening experience was built on this foundation. It was the visible deliverable that allowed business development to move at the pace the AUM growth required.
Staffing combined global, domestic, and local contractor networks with targeted direct hiring. The model traded the certainty of a single sourcing channel for the flexibility to ramp where the work was. The team scaled 10x over two years with 100% retention.
What the numbers actually mean
The $1.3B AUM growth (from $900M to $2.2B in twelve months) was the headline. Underneath it:
- Cycle times dropped 60%, which is the difference between weeks-to-ship and days-to-ship for a typical feature.
- Defect escape rates dropped 40%, which is the difference between an enterprise customer’s first impression of “buggy” and “reliable.”
- SOC2 Type 2 was achieved on schedule, which unlocked the segment of clients that had been gating further growth.
- Team retention stayed at 100% through a 10x scale. The transformation was a leverage multiplier for the engineers, not a threat.
What made it work
Three decisions, in order, made the rest possible.
The first was treating AI-SDLC as an operating-model change, not a tool adoption. Specifications, governance, context infrastructure, and human-in-the-loop gates were established before the agents were turned on at scale. Most companies do this in the opposite order.
The second was making the architecture board the governance body for the new model, rather than building a parallel organization. This kept decision-making authority where the codebase knowledge already lived and prevented the common failure mode of having “AI strategy” and “engineering reality” diverge.
The third was honest staffing. Mixed sourcing, transparent expectations, and an explicit retention strategy. Treating the transformation as a productivity multiplier rather than a cost-cutting exercise is what kept the team intact.
What is portable
Not every company will execute this transformation in twelve months. Not every company has a board that will fund three concurrent tracks. But the order of decisions, the governance-before-tools discipline, and the human-in-the-loop framing are portable to any company serious about agentic development at production scale.
The technology gets faster every quarter. The discipline takes time to grow.