Can AI Help Us Make Better Decisions?

Introducing What Do I Do: An AI Decision Support System

Every day, we make dozens of decisions. From trivial choices to life-changing ones, we are constantly faced with decisions that affect us and, sometimes, the people around us. Sometimes we try to run away from making these decisions because we think they are super complicated or we are simply afraid of making the wrong decisions. We try to outsource our decision making to friends, parents, superiors, mentors, random anonymous pages online, paid coaches, and recently, even AI.

One thing that I find very important in decision making is the way you think. It is not just about ensuring you make the right decision, but about asking the right questions that guide you to that “right” decision. I believe good decisions begin long before the final choice; begin with the questions we ask ourselves. And with the AI tools we have, it’s not about AI making decisions for us. It is about intelligently coming up with better questions to help us land at the right decisions.

Why decision making fascinates me

Why am I talking about all these?

I’ve always enjoyed helping people think through difficult decisions. Friends and family often come to me when they’re stuck, and I genuinely enjoy unpacking the problem with them. I avoid making decisions for them, so I walk through the different options with them and try to ask the right questions that help them arrive at their own decision. 

In the past few months, I’ve been thinking a lot about how we make decisions and how we can use AI to improve our decision-making process  by helping us think more clearly, uncover the questions, assumptions, and blind spots that lead to better decisions. 

This has led me to my first experiment (project) out of my personal technical lab, the place where I will explore things I find curious, try to solve problems I observe around me, build projects, make mistakes, fail, grow, learn, and the best part? Share it all with you.

The Problem

Imagine you’re deciding between two job offers. Traditionally, you would maybe go speak to a career coach, your parents, or a mentor in your industry, asking them what you should do. If you’re like me that loves a spreadsheet, you might create a spreadsheet with both job offers comparing different factors like 

• salary
• title
• benefits
• remote policy
• vacation

Eventually you pick one.

Six months later you realize you were comparing generic factors, without considering what really mattered to you and aligned with your values and goals. Maybe you didn’t consider

• commute time
• growth opportunities
• company stability
• alignment with your long-term goals

These are just a few factors that could drastically affect your decision but are easy to overlook. Missing them could lead to a decision you later regret.

The quality of decisions you make is not only based on your personal intelligence, but based on the information you have available to you at the time of making the decision, identifying the major points, assumptions, biases, constraints, implicit beliefs, facts, and asking the right questions to get the right information.

The Solution: “What Do I Do?”

Introducing “What Do I Do?”. I call the project What Do I Do because that’s the question most of us ask when we’re facing an important decision. My hope is that this becomes a personal support system whenever you’re faced with a decision or dilemma. I’m exploring what it means to use AI as a decision support system and reasoning guide, instead of just outsourcing critical thinking to AI.

Unlike a typical chatbot, What Do I Do? doesn’t immediately try to answer your question. Its first objective is to understand your situation. The system doesn’t immediately give advice; instead, it tries to understand how you think. Then it asks follow-up questions designed to uncover:

• your values
• constraints
• biases
• assumptions
• risks
• missing information

Only then does it generate recommendations. 

This project isn’t just about building an AI application. It is my first project focused on building reliable agentic systems that solve problems for people like me daily.

How It Works

The user experience for this agent is very simple. The user visits the web app, presents their dilemma, then the system automatically generates a sequence of questions for the user to answer. 

These questions aim to maximize information gain per iteration, extracting the user’s values, constraints, priorities, potential risks, and unspoken thoughts. 

It does not make a decision for the user. Instead, it helps you make a more informed decision by explaining its reasoning, presenting supporting evidence extracted from your answers, highlighting trade-offs, and expressing its confidence in the recommendation.

Project Goals

Once I get an MVP of this ready, my goal is to share it with you all to try out. I don’t want this to be just another AI app or agent, I’m approaching this from more of a systems thinking lens. I’m less interested in building a chatbot and more interested in building a reliable decision-making system.

The success criteria for this project is not how great the model is, or how fancy the technologies used are; I am more focused on how the system is designed and architected, how the agent is orchestrated, how the state objects are updated and information is maximized, and the reliability, observability and evaluation of the agentic system, all while keeping cost and latency low.

What I’ll Be Sharing

If you’re interested in building an agent, or curious about the architecture, system design, tradeoffs, and design decisions involved in agentic engineering, this is for you. I’ll be sharing periodic updates as I make progress on this, with not just the technical decisions, but also the business decisions, and my thought process in building this. These will not just be progress reports; rather I’ll be sharing what I’m learning as I build reliable AI systems and experimenting with different technologies in the agentic space. 

I’ll be documenting architecture decisions, evaluation frameworks, prompt iterations, failure modes, latency optimizations, and the trade-offs behind major design choices. I will also share heavily on instrumenting, testing and evaluating agentic systems, experimenting, running analysis on these, and implementing an iterative feedback loop to improve the product based on this. Basically putting on the Data Science hat that is highly needed in AI projects.

My hope is that by sharing both the successes and the mistakes, other data scientists and engineers can learn alongside me. This project is open source and you can also find the code on GitHub, so feel free to check it out. 

What Do I Do is still in its early stages, and I’m sure many of my assumptions will turn out to be wrong. That’s part of the reason I’m building it in public.

If you’re interested in AI systems, agentic engineering, or simply making better decisions, I hope you’ll follow along. Let’s see where this experiment takes us.