A platform engineer at a national bank needs to push a script across every machine in a server environment. The old method usually took a day or two: read the documentation, cross-check the configs, then write the deployment plan by hand.
This time, he fed the system context and the objective to an AI assistant, checked the output against vendor documentation, and had a working plan inside 90 minutes.
That scene comes from a workflow audit I ran with Dan, an infrastructure and systems engineer on a ten-person team at a regulated bank. His setup runs fully on-prem, because data sovereignty rules keep the bank’s data out of the cloud.
So every constraint that makes AI adoption harder applies to him: high uptime demands, tight privacy rules, and a stack he’s still learning after moving from one bank to another.
As suspected, my audit found that AI didn’t replace his judgment, but significantly compressed his research. The biggest savings came from planning unfamiliar work, not writing code.
He runs two tools for two jobs, and his employer enabled all of it through access first, training second, and one narrow privacy rule as part of governance.
AI case studies aren’t common in infrastructure engineering
Most published AI stories follow knowledge workers: marketers, analysts, executives. The people who keep banking infrastructure running rarely appear.
Yet across the financial sector, this is where adoption concentrates. Financial institutions use AI mostly for internal operations and regulatory compliance, with limited use for customer-facing or revenue tasks.
Dan’s team fits that pattern. They handle storage, the servers where applications live, and deployments other departments request. There are tight limits on what leaves the building, starting with a ban on client data in any AI tool.
The same lean toward internal use shows up by region: a regulator survey in Hong Kong found most banks there reaching for AI first in operational automation and document processing. Regulators worldwide point to the same risks behind the privacy rules—specifically naming model error, data governance, and dependence on a few providers.
Where the time savings come from
Before AI, troubleshooting meant a search across documentation pages, skimming articles that had nothing to do with the problem to reach the one that did. That’s the cost AI removed first. Instead of reading ten pages, he gets pointed to the fix immediately.
Deployment followed the same pattern, with a task that once took 1-2 days dropping to under 90 minutes. The AI didn’t write the script so much as structure the approach: how to stage the rollout across machines, given the system and constraints he described.
In a controlled study, developers with an AI pair programmer finished a coding task 55.8% faster than those without. Dan’s time saving runs larger because his slow step was planning the approach in an unfamiliar system, not typing the code.
Two AI tools for two jobs
Dan runs Microsoft Copilot for daily prompting and automation, and Google Gemini for checking sources. Gemini’s search mode returns links to legitimate vendor documentation, which lets him confirm a fix against the primary source before he runs it. Copilot doesn’t surface those docs by default, so it covers the answer but not the proof.
Dan’s distrust runs across the field: in Stack Overflow’s 2025 survey of more than 49,000 developers across 177 countries, 84% use or plan to use AI tools, while only about 33% trust the output to be accurate. For an engineer running changes in production at a bank, a tool that only answers isn’t enough.
What the bank got right
Three choices made the adoption work.
First, access before gatekeeping: Copilot reached every department through existing Microsoft licenses, on the paid tier instead of the basic one.
Second, the bank runs periodic refresher courses with quizzes, so staff who don’t understand AI get a route in.
Third, they kept governance narrow: no client data in AI tools, a few blocked applications, and a new CIO who set AI adoption as an objective, which gave existing use official backing.
The tools came first, the people who knew them started, and the policy caught up to the practice.
Banks are pouring money into this. Sector spending on generative AI alone is projected to climb from USD 3.86 billion in 2023 to almost USD 85 billion by 2030, most of it going to headcount and IT infrastructure.
What AI still doesn’t touch
AI left two things untouched in Dan’s day to day. The first is on-call standby. Dan still opens his laptop on weekends when a server issue escalates, and that remains the one part of the job he calls a drain. AI sped up his research, but did nothing for the on-call model, which is a staffing and process problem.
The second is the verification step. Even with AI, Dan double-checks, sometimes triple-checks, before he runs anything in production. The caution makes sense here, and it caps how dramatic the savings get on high-stakes work.
The wider data shows the same restraint, with developers resisting handing AI the riskiest, most systemic tasks, and 76% saying they don’t plan to use it for deployment and monitoring. Speed on the planning end doesn’t change the care needed at the deployment end.
What to take from this AI workflow case study
A few patterns from Dan’s audit carry over to anyone in a similar seat.
- Lead with research, not code. The compounding savings come from cutting the time spent finding the right answer in systems you don’t know yet.
- Match the tool to the task. General prompting and source-checked troubleshooting need different tools, and running both costs nothing.
- Keep governance narrow. One clear rule kept the door open without a heavy framework.
- Don’t read AI adoption as handing over control. This team adopted at peer level, with every member using the tools and a person checking every output before it reached production.
Not sure where AI actually fits in your workflow? A free thirty-minute audit maps your current process and points to the highest-value places to start.
Mo Shehu, Ph.D. writes and speaks, and consults on AI, digital strategy, and communications.