She Trained the Chatbot That Got Her Fired. Here's What Bankers Miss

Kathryn Sullivan spent 25 years at Commonwealth Bank of Australia. She'd worked branches across Brisbane and the Gold Coast, and when the bank moved her into online customer service, she embraced it — including the part where she developed scripts and tested responses for a new AI chatbot called Bumblebee. "I could see the benefits," she later said. "Except I didn't look far enough to see it taking my job."

In July 2025, she was given an hour's notice before a meeting that told her she was one of 45 employees being made redundant — in favour of the technology she'd helped train. The bank ultimately reversed the redundancies under union pressure. The chairman called it "a mistake." But when Sullivan confronted leadership at the October 2025 annual meeting, she noted that not all the jobs offered back were the same jobs. The reversal was real. The restoration was incomplete.

Kathryn's error wasn't ignorance of AI — she was literally building it. Her error was not understanding where the accountability boundary was supposed to sit. That distinction is now the most important thing to understand about what AI is doing to banking work.

This isn't a story about whether AI will change your job. It already is. The question worth asking is which part of your job, and what the disclosed corporate plans actually say about where the cuts are landing.

What AI Is Already Doing to Your Workload

The automation happening inside banking right now is concentrated in a specific, nameable category: tasks that involve assembling information, processing documents, generating first drafts, or handling routine customer requests. Banks have stopped piloting these capabilities. They're disclosing results and setting formal workforce targets around them.

Standard Chartered's May 2026 investor presentation set a formal target of more than 15% reduction in corporate-functions headcount by 2030. The bank defines those functions explicitly as non-frontline support services — document processing, back-office operations. The Guardian reported this translates to roughly 7,800 roles out of more than 52,000 in those functions. This is not an analyst forecast. It is a disclosed corporate plan.

TD Bank told investors on its May 2026 earnings call that agentic AI had reduced mortgage pre-adjudication from approximately 15 hours to three minutes. The process involves scanning client documents, calculating income, running consent checks, and generating a memo for underwriters — tasks that were previously human-assembled. TD was tracking ahead of a $200 million annual AI-value target at its fiscal-year halfway point.

HSBC is using generative AI to assist credit-analysis write-ups and support an institutional-servicing assistant that handles 3 million client interactions annually. Citi, at its May 2026 investor day, described compressing KYC, reconciliations, loan operations, and document processing "from months to days to minutes."

Meanwhile, the U.S. Bureau of Labor Statistics projects teller employment will fall 13% by 2035 — from 339,200 to 294,500 jobs — driven by digital banking, branch reduction, and automation. That's a 10-year forecast, not a single-year cliff. But the direction is unambiguous.

The agentic AI agent will summarise different information from different data points and provide the right information to a human.
— Sandra Aziz, Senior Machine Learning Engineer, TD Bank

The tasks in motion share a common structure: they involve retrieving, assembling, or generating information according to a known pattern. If a meaningful share of your current workload fits that description — compiling client files, drafting routine correspondence, processing applications, handling predictable customer queries — that work is being actively redesigned around you. Whether you're at a community bank or a global institution, the timeline matters more than the job title.

What's Actually Becoming More Valuable

The roles gaining value share a different structure from the ones contracting. They require a human to validate AI output, take accountability for a decision, explain it to another person, or handle the case that doesn't fit the pattern. That's not a soft description — it's a specific, learnable skill cluster. And the wage data confirms it's being priced accordingly.

Sandra Aziz joined TD Bank in 2014 as a statistics intern. Her team's name changed from "statistical team" to "analytics team" to an AI-development unit without her ever leaving the institution. By March 2025 she was leading development of TD's first agentic AI application — the same mortgage tool that now processes pre-adjudication in three minutes. Her observation captures where human work is still indispensable: "The agentic AI agent will summarise different information from different data points and provide the right information to a human."

She was working at the accountability layer — defining what the machine hands off to a person and what that person does with it. Kathryn Sullivan, by contrast, was working at the machine-facing end of her job: building what the bot knew. The distinction between those two positions is now worth real money. PwC's 2026 AI Jobs Barometer, analysing over one billion job postings, found that "professionalised" roles — where AI automates routine tasks so human judgment and expertise are emphasised — are seeing 42% faster salary growth than "democratised" roles, where AI makes the task itself easier for non-experts. In financial services specifically, the AI wage premium hit 53% in 2025. AI-related job postings in the sector surged 77.4% year-over-year, against 12.8% growth across all sector postings.

Bobby Grubert, Head of AI and Digital Innovation at RBC Capital Markets, describes what this looks like at the senior level: "AI helps transfer and scale institutional knowledge so junior team members contribute at a higher level sooner, while senior professionals focus on judgment-intensive work." The augmentation argument holds — but only if you've positioned yourself at the judgment layer, not the assembly layer.

This pattern holds across banking functions. In lending, it's the officer who reviews an AI-generated credit memo and explains an adverse decision to a borrower. In wealth management, it's the advisor who interprets a tool's recommendation in the context of a client's actual life. In compliance, it's the analyst who investigates an AI-flagged anomaly and decides whether it warrants escalation. The common thread is accountability, not a specific technical credential.

The Honest Picture: Both Things Are True at Once

Knowing what's contracting is only half the equation. The more urgent question — especially if you recognise your tasks in that list — is what the career trajectory looks like for people who moved toward accountability work. But there's something the augmentation story tends to skip: the banks betting biggest on AI aren't all expanding their workforces. Some are doing both at once, and understanding which is happening where matters for how you position yourself.

Citi's May 2026 investor day deployed AI tools to more than 180,000 colleagues while simultaneously planning to add more than 400 client advisors and personal bankers and more than 200 small-business advisors. The bank reported that customers receiving dedicated advisory coverage generated nearly 2.5 times the investment revenue of those without it. AI-in-operations and relationship-hiring are running in parallel at the same institution.

The biggest opportunity is the democratization of expertise — AI helps transfer and scale institutional knowledge so junior team members contribute at a higher level sooner, while senior professionals focus on judgment-intensive work.
— Bobby Grubert, Head of AI and Digital Innovation, RBC Capital Markets

New York Fed business surveys from August 2026 offer a more granular check on the layoff narrative. Among regional service firms using AI, only 4% reported AI-related layoffs in the prior six months. But 15% hired fewer workers than they otherwise would have — a quieter, slower-moving form of displacement that doesn't generate headlines. It compounds over time through unfilled vacancies rather than announced cuts.

The Cambridge Centre for Alternative Finance's April 2026 global financial-services survey found 24% of industry respondents expected net role reductions, while 44% saw commercial and wholesale banking as likely to see net job increases. The gap isn't a contradiction — it reflects a real split between functions. Support, document-processing, and routine customer-service roles are contracting through attrition and explicit targets. Relationship, specialist, and accountability-layer roles are growing at some institutions. Which side of that gap your role sits on is a question about your specific function and employer, not about banking in general.

The relevant diagnostic isn't "is my industry safe?" It's what your employer is publicly saying about which functions are being reshaped. Disclosed investor-day plans, transformation presentations, and official redundancy announcements are more reliable signals than industry averages.

What You Can Actually Do This Week

Which brings us back to the question Kathryn Sullivan's story leaves open: now that you know where the boundary sits, what can you do about it?

Kathryn understood Bumblebee's value better than most people in the bank. What she didn't have was a clear answer to one question: "In this new arrangement, where does the human have to be?" She was working at the machine-learning end of the job. The reviewers, the exception-handlers, the explainers — those were the roles that persisted. That gap isn't obvious until it isn't, and it closes faster than you'd expect.

The split happening inside banking right now isn't between AI users and non-users. It's between people who understand where accountability sits in their workflow — and have made themselves indispensable at that boundary — and people who haven't asked the question. That's a positioning decision, not a technical one.

Three moves, startable this week:

Audit your task list for assembly work. Write down every recurring task in your current role and mark any that primarily involve retrieving, compiling, or formatting information according to a known pattern. Those specific tasks are in motion. Knowing which ones they are is the first step to moving away from them deliberately.

Find your accountability moment. In your current workflow, identify the point where a human has to sign off, explain a decision to a customer or regulator, or handle a case that doesn't fit the standard process. That's your leverage point. Invest time there, document your decisions, and make your judgment visible.

Ask your employer a direct question. Request clarity on which AI tools are being deployed in your function and on what timeline. If your institution has published a transformation presentation or investor-day materials, read them. The language about which functions are being "simplified" and which are being "expanded" is usually specific.

Kathryn's story wasn't inevitable. It was a failure of positioning — one that's still avoidable, for most banking roles, if the question gets asked before the meeting does.

Recommended Tools & Resources