Agentic AI Reply Student Clash, Imperial, 2026

A news agent that teaches you finance while you read

One day to build and pitch an agentic AI concept for financial services. The hard part was not thinking of ideas. It was committing to one.

The problem we picked

Financial news is written for people who already understand it. Everyone else scrolls headlines, absorbs the sentiment and learns nothing. The gap is not access to news. It is that nobody explains the vocabulary at the moment it is needed.

Step 01
Calibrate
A short quiz establishes financial literacy, whether the reader is a student or a professional, and which topics they care about: inflation, jobs, housing, markets.
Step 02
Select and summarise
The agent pulls from trusted public sources, picks the stories that matter for that reader, and writes each one in two or three lines of plain language.
Step 03
Test at the point of confusion
After each story, one small question. If a term like base rate or yield is not understood, it is explained immediately with a short real-life example.
Step 04
Adapt
The agent decides what to show next, how hard to make it, and which concepts to revisit. Retrieval-augmented generation keeps every explanation tied to a real article or a vetted glossary, which is what holds hallucination down.

The product is deliberately narrow. It reads the news for you and teaches you finance in the background. It gives no investment advice, and that boundary is a design decision rather than a disclaimer.

What the day taught me

There is no perfect idea. The hardest part is not generating one, it is committing. We chose a direction, held it, and improved each iteration instead of restarting every time something looked more interesting.

Execution beat completeness. A simple narrative of problem, agentic solution, impact and governance landed far better than trying to show everything the technology could do.

And agentic AI matters in financial services for a specific reason. It is not question answering. It is systems that sense, decide and act inside strong guardrails: monitoring risk, educating customers in real time, and escalating to a human when the stakes require one.

The takeaway

Commitment is the scarce resource, not ideas.

Team

A team of four at the Reply Student Clash on agentic AI, hosted at Imperial. I worked on the problem framing, the governance boundary and the pitch narrative.

All writing