PenFed Credit Union is one of the largest credit unions in the United States, with roughly three million members and a balance sheet that sits in the top tier of the cooperative banking sector. Its CEO, James Schenck, has been in financial services for twenty-five years and has seen multiple technology cycles arrive with great noise and leave with modest results. So when he tells diginomica that the credit union's Salesforce Agentforce partnership has been central to navigating the rate volatility and credit pressure of 2025 and early 2026, the claim deserves attention.
The substantive argument Schenck makes is that Agentforce was not deployed as a customer service chatbot. It was deployed as an operating-model intervention. Member-facing agents handle a growing share of inbound contacts, but the more important shift is in the back office, where agents handle exception flows in lending operations, deposit servicing, and fraud triage. Each of those domains has historically required experienced human operators who are expensive to hire and slow to train. The agents do not replace those operators, but they handle the routine variants and free the humans for the cases that genuinely need judgment.
For PenFed, the timing mattered. The 2025-2026 rate environment has been brutal for community lenders. Deposit competition has driven up funding costs, credit losses have ticked up in unsecured lending, and members have become more demanding about service. Without the operating leverage from Agentforce, Schenck argues, the credit union would have needed to choose between hiring through the downturn or accepting service degradation. The agents allowed it to do neither.
The conditions that made this work are worth itemizing for CTOs and VPEs evaluating similar moves. First, data foundations came before agents. PenFed invested in Data Cloud and unified member records well before the agentic layer arrived. Agents that cannot see a complete member view produce embarrassing answers. Agents with a unified view produce useful ones. Second, regulatory engagement was treated as a parallel design input, not a late-stage approval gate. The credit union worked with the NCUA and its internal compliance function to define what agents could and could not do, how decisions would be logged, and how human review would be triggered. Third, the change management investment was substantial. Frontline staff had to learn to work alongside agents, to challenge agent outputs, and to escalate when something looked wrong. The training program ran for months and is still ongoing.
For European banking clients considering similar deployments, the regulatory translation is non-trivial. EU AI Act obligations for high-risk financial services use cases are stricter than the US framework, and the documentation burden is heavier. But the underlying playbook holds. Data foundation first, regulator engagement in parallel, change management as a real workstream, and a clear definition of what success looks like in service quality, operating cost, and risk metrics.
The competitive implication for the broader Salesforce ecosystem is also clear. Agentforce has had a mixed first year in market, with strong logo wins offset by slower-than-expected revenue ramp. References like PenFed matter disproportionately because they are credible, regulated, and prepared to discuss specifics. Expect Salesforce to lean hard on this case in the next round of banking sales cycles.
There is a cautionary subtext as well. Schenck is candid that the deployment was harder than the marketing materials suggested. Data quality issues that looked manageable in pilot became blocking in production. Some agent capabilities that worked in demo environments produced unacceptable error rates against real member data. The credit union retrenched, fixed the data, and retried. That pattern, retrench and retry, is the honest experience of agentic AI in regulated industries right now, and any CTO presenting a smoother narrative to a board is either lucky or selectively reporting.
For bruno.digital's banking practice, the takeaways are concrete. Use PenFed as a reference in conversations with mid-tier European lenders evaluating Agentforce or competing platforms. Insist on data foundation work as a prerequisite, not a parallel stream. Treat regulator engagement as a design input. Budget realistically for change management. And expect the production deployment to be harder than the pilot suggested, because it always is.
One additional dimension worth flagging is the Industry 4.0 read-across. The same operating pattern PenFed is running, agents handling routine variance while humans handle judgment cases, is now showing up in manufacturing operations control rooms, in logistics dispatch centers, and in field service planning. The technologies differ, but the change management challenge is identical: frontline operators have to learn a new partnership model with software that acts rather than reports. The organizations that move first on that cultural shift will compound advantage. The organizations that try to install the technology and skip the cultural work will produce expensive disappointment and, more dangerously, a generation of operators who lose trust in AI and resist the next wave.
