Book a demo
Topics · AI-DLC

Asset Mgmt AI-DLC Challenges | Moring

Asset managers adopting AI-DLC face a widening gap: AI has entered investment processes faster than governance has kept pace. Challenges include code generated outside institutional control, model risk expectations built for human-led development, lifecycle checkpoints that traditional review cannot staff, and boards asked to oversee technology they cannot yet see.

On this page
  1. Short answer
  2. How it works
  3. FAQs
  4. Related reading
  5. Sources

Short answer

Asset managers adopting AI-DLC face a widening gap: AI has entered investment processes faster than governance has kept pace. Challenges include code generated outside institutional control, model risk expectations built for human-led development, lifecycle checkpoints that traditional review cannot staff, and boards asked to oversee technology they cannot yet see.

How it works

Advisory research on asset management is blunt about the core challenge: AI has moved from experimentation into early production across investment processes, but governance has not kept pace with adoption. AI-DLC amplifies that gap because it changes not just tooling but the shape of work. The AI-driven development lifecycle positions AI as the primary executor across planning, coding, and operations, with humans directing, approving, and overseeing, and methodologies built for human pace are strained when generation happens in minutes and review capacity stays constant.

The most concrete challenge is control. Financial-institution analysis of AI-DLC adoption observes that institutions are adopting AI coding agents at pace, but most lack a governance framework that satisfies regulators or their own risk appetite, producing powerful tools that generate code outside institutional control. For an asset manager, uncontrolled generated code reaching portfolio systems, order management integration, or client reporting is a model risk problem in substance even though it arrives dressed as a productivity win, and model risk management expectations were designed around models, not agents writing software.

A second cluster of challenges is structural. AI-DLC compresses development cycles and shifts security validation earlier, which means control points designed for quarterly releases no longer align with delivery reality. Individual engineers gain speed quickly, but delivery does not automatically accelerate, because constraints migrate to reviews, product decisions, and context maintenance. Asset managers adopting AI-DLC therefore discover their bottlenecks relocate: approval gates and review throughput become the limiting factor, and understaffing them tempts teams to skip the very controls the lifecycle requires.

A third challenge is oversight credibility. Corporate governance commentary argues AI is a board-level governance issue requiring compliance discipline and operational oversight, not merely a technology deployment question. Asset management boards already carry model risk, valuation, and fiduciary duties; AI-DLC adds questions boards are poorly equipped to ask: which agents wrote what code, who approved it, and what evidence exists. Reporting that models degrade, drift, and accumulate regulatory debt without structured oversight applies with extra force when the agents themselves evolve weekly.

These challenges are tractable, but only through architecture rather than aspiration. The remedy set is consistent across the research: named lifecycle stages with structured records, per-action approval gates proportionate to consequence, evidence retention that makes any change reconstructable, and cost observability so the economics of agent-driven development are visible. Model risk management traditions within the firm, familiar from SR 11-7-era practice and its successor SR 26-2 expectations, supply the cultural template: independent review, documented validation, and clear accountability extended to a lifecycle where much of the code is machine-authored.

Frequently asked questions

What is the biggest challenge asset managers face adopting AI-DLC?

Control catching up with adoption. Advisory research on asset management notes governance has not kept pace while AI entered production across investment processes, and financial-institution analysis of AI-DLC observes most firms lack a governance framework satisfying regulators or their own risk appetite, leaving powerful tools generating code outside institutional control. For an asset manager, generated code reaching portfolio or client-reporting systems is a model risk concern regardless of productivity gains. The fix is structural: named lifecycle stages, approval gates, and evidence that makes every machine-authored change reconstructable.

How does AI-DLC change where asset managers feel delivery bottlenecks?

Those constraints migrate downstream of generation. Research on AI-DLC practice observes that individual engineers get faster quickly, but delivery does not automatically accelerate because constraints shift into the surrounding system: reviews must keep pace with generated changes, product decisions must not lag, and context must stay current. Asset managers feel this as review and approval throughput becoming the binding constraint, which pressures teams to weaken controls exactly when consequence is rising. Treating review capacity and approval gate design as first-class investments prevents that erosion.

Why is AI-DLC a board-level issue for asset managers?

Because it changes what the board is implicitly certifying. Corporate governance commentary frames AI as requiring board attention, compliance discipline, and operational oversight rather than purely technical management. Asset management boards already answer for model risk, valuation integrity, and fiduciary duty; AI-DLC adds machine-authored code entering those systems and agents whose behavior evolves rapidly. Directors need answers to which agents produced what changes, under what approvals, with what evidence, and lifecycle governance that produces those answers is what converts AI-DLC from an unquantified exposure into a managed one.

How do model risk traditions help with AI-DLC challenges?

They supply the cultural template. Model risk management as practiced under SR 11-7 and continued under SR 26-2 rests on independent review, documented validation, clear accountability, and evidence suited to regulatory examination. AI-DLC challenges map directly: machine-authored code needs independent review and validation evidence just as models do, agents need named ownership, and lifecycle stages need structured records. Extending these habits to the AI-driven lifecycle lets asset managers govern novel tooling with muscles regulators and internal audit already recognize, rather than inventing governance from scratch mid-deployment.

Sources

  1. - federalreserve.gov
  2. - federalreserve.gov
  3. - artificialintelligenceact.eu