Banks have spent a decade buying agile and getting ceremony instead. Standups, story points, quarterly planning theatre, a coach in every department, and delivery cycles that still move at the pace of the annual budget. David Fapohunda, who works in delivery within financial services, argues that most of what institutions call agile transformation never touched the element that determines speed: who owns the outcome and where the money sits. That distinction is about to get too expensive to ignore, because AI is dismantling the assumption agile was built on. Sprints exist to coordinate humans. When a coding agent produces a credible first pass in minutes, coordination stops being the bottleneck and governance replaces it.
Funding Models, Not Ceremonies
The clearest evidence that a bank has done the work is structural, not behavioral. Fapohunda points to ING’s rebuild, where roughly 3,500 people moved into about 350 nine-person squads inside 13 tribes, each tribe capped near 150 people, a deliberate nod to Dunbar’s number. What that produces on the ground is one persistent team owning a payments or mortgage product end-to-end, shipping in weeks rather than quarters. “It is not standups and story points. It is a structural change in who owns outcomes,” he says. McKinsey finds banks that bring agile into the technology function capture 15 to 20 percent productivity gains and can lift return on equity by 1 to 1.5 points, but the mechanism is ownership, not method.
This is also the honest answer to the critics. When commentators say agile has failed at scale, and when McKinsey itself points banks toward a “product operating model,” Fapohunda reads it not as the death of agility but as agility without the theatre. “The industrial version is dying, and rightly. What survives is the original idea: small teams that own a product, ship often, and answer for outcomes.” The question was never whether to hold ceremonies. It was whether anyone owned the result.
That framing explains why so many programs stall. Fapohunda’s read is that the obstacles are almost never technical. In the 16th and 17th State of Agile Reports, 47 percent of practitioners name organizational resistance to change as the top barrier and 41 percent cite insufficient leadership participation; KPMG’s Survey on Agility found 75 percent of firms lack the cultural support their agile maturity assumes. Banking adds a harder constraint underneath the cultural one. “You cannot run durable product teams on annual project funding and functional silos,” he says. The fix is unglamorous and political: move money to persistent teams, collapse management layers, and make a named senior leader personally accountable. Delegating the change to a coach is how institutions buy the vocabulary and keep the operating model.
Compliance as a Design Input
The standard objection in regulated delivery is that rigour and speed trade off against each other. Fapohunda rejects the framing outright. “Stop treating compliance as the brake. Treat it as a design input, because regulators now expect exactly what mature agile produces.” DORA, in force across the EU since January 2025, mandates tested change controls, ICT risk management, and demonstrable operational resilience. Read closely, those are delivery engineering requirements, not documentation requirements. They describe a pipeline, and a bank that has built one is closer to compliant than a bank that has built a control committee.
The practical consequence is where the controls live. Build them into the delivery pipeline rather than bolting them on afterward: automated testing, separation of duties enforced in code, and audit evidence generated by the deployment itself. Done that way, the risk profile inverts. “Agile makes you more compliant, because every change is small, logged, and reversible, unlike a big bang release nobody can unwind,” Fapohunda says. That is the argument compliance functions in banking have been slowest to accept, and the one with the most leverage in it. A quarterly release is a single large unrecoverable event dressed up as caution. A small reversible change is the safer instrument, and it happens to be the faster one.
Agility Has to Move a Number
Fapohunda is impatient with agile discussed as a virtue rather than a lever. “It only earns its place when it changes a number a CFO cares about: cost to serve, loss rates, time to market.” He speaks from direct experience leading fraud operations at scale, where embedding faster, iterative delivery into fraud controls moved loss rates directly, because the team tuned models and rules in near real time instead of waiting for the next release cycle. That is the test he applies to any delivery model. “If it cannot move cost to serve, loss rates, or time to market, it is process for its own sake.” The measurable outcome, not the maturity assessment, is the only scorecard that survives a budget review.
From Coordinating People to Governing Agents
The unit of delivery is changing under the industry’s feet, and Fapohunda is blunt that existing agile mechanics were not designed for what is arriving. McKinsey reports that AI-led organizations have seen 20 to 40 percent reductions in operating costs and 12 to 14 point gains in EBITDA margin. As agents take the first pass at code, the engineer’s job shifts to delegate, review, and own. “Agile has to evolve from coordinating people to governing agents: smaller teams, more senior judgment per person, and controls that assume a non-human contributor is committing work.” Note what that implies for team design. The nine-person squad was sized for human throughput. Governance of machine output is a different problem with a different headcount.
Three moves follow, and the sequencing matters more than the list. First, retire project funding for anything digital and fund durable teams instead; technology spend in banking is rising roughly 9 percent a year against revenue growth of about 4 percent, and Fapohunda’s view is that the gap closes through ownership rather than more headcount. Second, put governance into the pipeline before agents scale, or velocity becomes your single largest operational risk, a warning worth reading twice, because the failure mode is not slow delivery but fast delivery that no one can evidence. Third, retrain for judgment. “As agents take the first pass, the scarce resource becomes people who can review and own the result, not people who can produce it.”
Banks have spent years optimizing for production capacity. The institutions that fare best in the next cycle will be the ones that noticed early that the constraint moved to review, and rebuilt their teams, their funding, and their controls around that.
Follow David Fapohunda on LinkedIn for more insights on agile delivery, technology governance, and operating models in financial services.