Building with AI: A Journal of Learning, Shipping, and Figuring It Out

I’ve spent most of my career somewhere between data, technology, and products.

I started as a software engineer. Then I moved into data and analytics, working with increasingly large and messy datasets. At my previous role, I eventually found myself building products around those systems: customer analytics, CDP, recommendation engine, serving millions of customers.

Then I went to MIT Sloan.

And somehow, I found myself back at the beginning.

Not because technology had gone backward, but because AI was changing what it meant to build software in the first place.

At MIT, I started exploring generative AI, AI agents, RAG, MCP, A2A, agent orchestration, and the emerging infrastructure around agentic systems. I experimented with building agents myself through AI Studio Course.

The more I learned, the more I realized that I didn’t want to simply keep up with AI.

I wanted to understand what it means to build with it.

Why this journal exists

There is no shortage of AI content on the internet.

Every day brings another model, framework, benchmark, agent platform, prompt technique, or prediction about the future.

I’m not particularly interested in adding to that noise.

Instead, I want to document something more personal:

What does it actually feel like to build products and make decisions in the middle of an AI transition?

I want to write about the things I’m learning, the things I’m building, and — importantly — the things I get wrong.

Some of these posts will be technical.

Some will be about product management.

Some will be experiments.

Some will probably be half-formed ideas that I haven’t figured out yet.

That’s intentional.

I want this journal to be a record of the journey rather than a collection of polished conclusions.

From Analytics to Intelligence

One idea I’ve been thinking about a lot is the difference between prediction and decision-making.

For years, much of my work revolved around questions like:

Who is likely to churn?

Which customer is likely to buy this product?

Which offer should we show this person?

Machine learning became very good at helping us answer those questions.

But knowing what is likely to happen isn’t the same as knowing what we should do.

That distinction has become increasingly interesting to me.

An AI system can predict.

It can recommend.

It can generate.

And increasingly, it can act.

But the real product challenge is figuring out when, why, and under what constraints it should act.

That’s where my interest in AI increasingly intersects with something broader:

Decision Intelligence.

How do we build systems that help people — or other AI systems — make better decisions under uncertainty?

How do we combine prediction, causality, experimentation, context, incentives, and human judgment?

And as AI agents become capable of taking actions themselves, how do we make those decisions trustworthy?

I don’t have all the answers.

That’s exactly why I want to explore them here.

What I’ll be exploring

This journal will probably revolve around a few questions:

1. How do we build useful AI products?

Not demos. Not impressive prototypes.

Products that actually solve problems, fit into workflows, and create measurable value.

2. How does product management change when the product can think and act?

Traditional software gives users tools.

AI increasingly gives users collaborators, assistants, and autonomous agents.

That changes product discovery, UX, evaluation, economics, and even what an MVP looks like.

3. How do we make AI systems trustworthy?

As agents gain more autonomy, questions around identity, reputation, permissions, context, governance, and accountability become product problems — not just technical problems.

4. What happens to decision-making?

I’m particularly interested in the space between analytics and action: how data, ML, AI, experimentation, and human judgment can come together to improve decisions.

5. What can we actually build today?

I’ll share experiments, prototypes, architectures, failures, and lessons whenever they’re useful.

Sometimes the best way to understand an idea is simply to try building it.

Learning in public

I’ve spent enough time in technology to know that every “overnight” breakthrough usually sits on top of years of accumulated infrastructure, experimentation, and lessons from things that didn’t work.

My own AI journey is no different.

The NBO system I built didn’t start with a sophisticated AI stack. It started with customer data, business questions, imperfect infrastructure, experimentation, and a very simple question:

Can we make the next offer more relevant to the customer?

Years later, I’m asking a much broader version of that question:

Can we build intelligent systems that help us make better decisions?

The technology has changed dramatically.

The fundamental product question hasn’t.

Build something useful.

Measure it.

Learn.

Improve it.

Repeat.

So this is where I’m starting.

Not as an AI expert who has figured everything out.

But as a product builder who has spent years working with data, analytics, and large-scale systems — and who now wants to figure out what comes next.

I’ll document the journey here.

Welcome to the journal.