How do you design metacognition into novice investors?
AI Investment Agent: a working conversational prototype that coaches novice investors through real decisions, built to sharpen judgment rather than replace it.
The Challenge
Most investing platforms focus on information delivery, yet many novice investors still struggle to make sound investment decisions.
Through research, our team found that the problem wasn't access to information, it was judgment and a lack of metacognition. New investors often don't know what they don't know. Some avoid investing because they lack confidence, while others become overconfident without recognizing gaps in their understanding.
We wanted to explore a different role for AI: helping users develop better decision-making skills rather than simply providing answers.
The Insight
Most AI products optimize for speed and efficiency.
In investing, faster answers don't necessarily produce better decisions.
Our research suggested that effective investing depends on metacognition: the ability to accurately evaluate one's own knowledge, confidence, and reasoning. Rather than teaching investing concepts alone, we focused on helping users develop better judgment under uncertainty.
Key Product Decisions
- Build for calibration, not information delivery
Users rated their confidence before receiving guidance, allowing the system to surface patterns of overconfidence and underconfidence over time.
- Make reasoning visible
Instead of evaluating outcomes alone, the agent prompted users to explain their thinking before receiving feedback.
- Model expert thinking
The agent was designed to model how experienced investors evaluate uncertainty rather than provide direct recommendations.
- Create low-stakes practice
Scenario-based exercises let users practice decision-making and reflection before risking real money.
How the Agent Works
We built a working prototype that users could converse with. The agent is not an information tool, it is a metacognitive coach. Before offering guidance on a decision, it asks the user to rate their confidence and explain their reasoning, then responds based on how that confidence lines up with their actual accuracy.
That calibration model is the core of the agent. It treats the gap between how confident a user feels and how accurate they are as the most important signal, and responds differently to each case:
"An experienced investor would pause here and ask: what downside risk am I accepting, and how does this choice fit my time horizon? Notice how this step comes before choosing specific assets."
It guides each user through a structured decision lifecycle:
This workflow mirrors real-world financial decision processes used in institutional investment environments.
Outcome
We built a working prototype that users could converse with, running the full decision workflow from onboarding through feedback. Rather than delivering answers, the agent improved confidence calibration, reflective thinking, and self-awareness, supporting better decisions instead of making them.
What I Learned
Many AI teams ask, "What can the model do?" A more important question is often:
"What type of thinking should it enhance for the human?"
Through this project, I learned that the most valuable role for AI isn't always generating answers. In high-stakes domains like investing, it can be helping users build better judgment, recognize gaps in their understanding, and make more informed decisions with confidence.