UBS is requiring AI proficiency for all graduates and interns in
When you look at the broader landscape, this is a pattern across the sector. Santander is already actively sourcing "advanced AI users," and Morgan Stanley has put out a projection that over 200,000 banking positions across Europe could be phased out within the next five years. We are seeing a transition where the "analyst" role is being redefined. The job isn't just about crunching numbers in Excel anymore; it's about who can orchestrate an LLM agent to automate that crunching and then provide the high-level synthesis.
How to actually prep for this shift
If you're trying to pivot into these roles or are currently a student, just saying "I use ChatGPT" won't cut it in a 2027 interview. You need a real-world deployment story. I'd suggest focusing on these three areas to build a portfolio that actually impresses a hiring manager:
1. Automating Data Extraction
Stop manually copying data from PDFs into spreadsheets. Learn how to use Python with libraries like PyPDF2 or langchain to build a RAG (Retrieval-Augmented Generation) pipeline that queries financial reports. If you can show an interviewer a local script that summarizes 10-K filings in seconds, you've proven efficiency.
2. Advanced Prompt Engineering for Analysis
Move beyond simple chat. Start using chain-of-thought prompting to break down complex financial models. For example, instead of asking for a summary, prompt the model to "Analyze the quarterly revenue growth of Company X, identify the three primary drivers of the increase, and cross-reference these with the macroeconomic trends in the provided PDF."
3. Building Small-Scale Agents
Try creating a basic agentic workflow. Use a framework like CrewAI or AutoGen to set up a "Researcher" agent and a "Critic" agent. Have the researcher find market data and the critic challenge the assumptions. Being able to explain the logic behind an agentic loop is exactly the kind of "AI skill" these banks are hunting for.
The reality of the job market
The Morgan Stanley figure of 200,000 lost jobs is sobering, but it's misleading if you think it's just about "replacement." It's about the baseline of productivity. A single analyst equipped with a custom AI workflow can now do the work of three junior staff members. The "entry-level" bar is simply being raised.
The key is moving from a "user" to a "builder." If you can't explain how you've optimized a prompt to reduce hallucinations in a financial forecast or how you've integrated an API to pull real-time ticker data into a LLM context window, you're just another user. To survive this transition, you need a hands-on guide to the actual tooling—Python, API integration, and vector databases—rather than just knowing how to write a decent prompt.