Risks and Opportunities

Section 11 of Agentic Banking Architecture: A Practitioner's Guide

A board member asked me at lunch recently what I see as the key risks and opportunities with agentic right now.

I gave three answers. On reflection, there's a fourth.

Most of the things commonly cited as agentic risks — control plane maturity, interface design, vendor lock-in, pace of adoption — are familiar problems wearing new clothes. They're real, but they're not structurally new. Banks have managed timing risks, design questions, and supplier concentration for decades. The substance is different, the management discipline isn't.

The four below pass a stricter test: they are structurally different about agentic transformation, not familiar problems in a new domain. Each is simultaneously a risk and an opportunity — the same asymmetry that runs through the rest of this guide. Banks that build the discipline to manage them capture compounding returns. Banks that don't pay in incidents, in foregone benefits, or in transformation that never actually happens.

1. Getting the deterministic-probabilistic boundary right

Treat a deterministic process as adaptive and you get variance where you need precision. Treat an adaptive process as deterministic and you forfeit the value agents bring. The boundary moves as organisations learn — what looks adaptive today often resolves into rules once patterns become clear, and what looks deterministic today sometimes needs judgement once the environment shifts.

The discipline is to manage the boundary explicitly rather than letting it default. Section 6 proposes one framework — three modes from predictable to adaptive, with different governance for each. The opportunity is the compounding return banks capture when they get the boundary right and revisit it as conditions change. The risk is paying in either direction: incidents where deterministic was required, foregone benefits where adaptive was warranted.

2. Codifying tacit knowledge

Agents work off explicit documentation. A lot of what makes banking run is in people's heads. If that knowledge isn't codified as transformation progresses, agents will make mistakes humans would not have made. Worst case, the knowledge is lost when people move on.

The codification work isn't preparation for agentic deployment. It is the deployment. The team that writes down its judgement calls, exception patterns, and informal escalation paths builds an asset that compounds across every agent that subsequently uses it. The team that doesn't builds agents that are confidently wrong. This is the activity-level point from Section 10 stated as a strategic risk.

3. Aligning organisational transformation with the technology

Banks that stay in silos, with each leader defending their fiefdom, won't transform. They'll automate existing processes inside existing boundaries. Significant investment, sophisticated agents, no transformation. No recovery of integration complexity. Strategic position unchanged.

This is the most significant of the four. The others become moot if the organisation can't transform across silos. Conway's Law — covered in Section 4 — is the structural reason previous integration waves (SOA, APIs, microservices) failed to dissolve the silos. Agentic technology won't dissolve them either if the organisational boundaries that created the silos are left intact. The opportunity is the bank that does both transformations together. The risk is the bank that does only the technology one.

4. Managing the human-agent boundary as a moving variable

In prior waves the human-machine boundary was set at design time and stayed there. With agents it moves continuously — as models improve, as confidence builds, as edge cases get codified. A transformation that required human judgement last quarter becomes automatable this quarter. A process that ran autonomously may need human re-engagement as the environment shifts.

Banks that treat the boundary as a one-time decision will keep finding themselves out of position. The capability that matters is the discipline to manage it as a moving variable — measured, calibrated, and revisited. The opportunity is the compounding value of capturing each shift as it becomes available. The risk is being structurally late to every move.

The test

All four are structurally new. They aren't familiar problems wearing new clothes. That's what makes them worth treating as the substantive content of agentic transformation rather than as items on a risk register.

This section is based on a LinkedIn post on the key risks and opportunities in agentic transformation, which originated in a board member's question.