From Writing Emails to Building AI Agents: My Journey as a Non-Tech Builder — Signal — Currly

In 2023, when ChatGPT became mainstream, I started using it for something embarrassingly simple.

Emails.

Event invitations. Follow-ups. PPT outlines. Content copies. Repetitive tasks that would normally take me two hours suddenly took thirty minutes.

Back then, this wasn’t considered impressive. In fact, many people disliked the idea. There was a subtle judgment around AI users.

“If you’re using AI, maybe you don’t know how to do the work yourself.”

Some believed it was cheating. Others simply didn’t understand it. I never argued with them. Because I wasn’t using AI to avoid work.

I was using it to think faster.

The Early Days: AI as an Assistant

Initially, AI was just a productivity tool for me. I would ask questions endlessly.

“How do I improve this email?”

“How do I structure this presentation?”

“Can you give me ten versions of this campaign copy?”

I wasn’t looking for magic. I was looking for leverage. And even in those early days, I could see something important:

AI wasn’t replacing my thinking.
It was amplifying it.

Then I Discovered I Could Build

My career was in marketing and partnerships. I was never interested in software engineering. Throughout college, I never imagined myself building applications.

That changed in 2024. I started experimenting with Claude and later the Anthropic ecosystem. For the first time, I could describe something in plain English and see it come alive.

My first project was ridiculous. I built a dashboard for my PUBG squad. I wanted to track kills, damage, rankings, and keep the competition alive among friends. The image processing was terrible. The app failed miserably.

But I learned something far more important:

I didn’t need to be a developer to build. I could create working prototypes.

And that realization changed everything.

Curiosity Became Obsession

I started building random things. Poker practice tools because I was terrible at poker. Small games. Dashboards. Experiments. Google AI Studio became my playground.

Every day I was asking: “What else can I make?”

The joy wasn’t in the final product. The joy was in watching ideas materialize from prompts. For someone who never wrote software professionally, that feeling was addictive.

The Shift From Prompting to Systems

Then I made another discovery. The quality of AI isn’t just about prompts.

It’s about context.

The same model behaves completely differently depending on the information you give it. I stopped obsessing over prompts. I started obsessing over:

01Context files

02Structured instructions

03Memory systems

04Skills

05Agent orchestration

06Tool usage

07Multi-agent workflows

This was no longer:

“Write me an email.”

This became:

“How do I create an AI that thinks the way I work?”

I Built My Own AI Team

In early 2025, when people had just started discussing AI agents, I built my own team of agents. Not theoretical agents. Working ones.

01One understood my campaigns.

02One knew my presentations.

03One followed my brand guidelines.

04One wrote emails in my tone.

05One worked on research.

I trained them using years of my own work:

My campaigns.My decks.My writing.My frameworks.My thinking style.

I remember posting on LinkedIn: “I have an AI team working alongside me.” At that time, it sounded crazy. Today, it feels normal.

I Burned Money to Learn

When I first started using Claude Code seriously, I burned through nearly $100 in less than two days.

—I was experimenting recklessly.

—Building apps.

—Testing workflows.

—Breaking things.

—Repeating the process.

But slowly, I learned:

01Skills.

02Environment management.

03Agent harnessing.

04Task decomposition.

05Parallel execution.

06Efficient context design.

The result?

70%

My token usage dropped by nearly 70%.

The same work became faster.Cheaper.More reliable.

Today, I Work Differently

Today, my workspace looks very different.

Multiple windows.Multiple agents.Parallel tasks.Dedicated contexts.Shared memories.Specialized skills.

I no longer think:

“How do I do this?”

I think:

“Which agent should do this?”

That’s a very different way of working.

The Biggest Transformation Wasn’t Technical

The biggest change wasn’t learning prompts. Or agents. Or code.

It was changing my identity.

I spent years believing: “I’m not a technical person.” AI challenged that belief.

I’m still a marketer.

But now I build.I experiment.I ship products.I design systems.

And honestly, that’s something I never expected from myself.

My View on AI Today

I don’t think AI will replace everyone. And I don’t think the hype alone will sustain it. But I strongly believe this:

People who know how to collaborate with AI— who know how to provide context, build systems, orchestrate agents, and think clearly— will operate at a completely different level.

The gap won’t be between humans and AI. It will be between people who know how to work with AI and people who don’t.

And I consider myself fortunate. Because I started with something as simple as rewriting emails. And somehow, that journey turned me from a marketer into a builder.

Rohit Bangaram · Founder, Currly

Currly started the same way: a problem, asked in plain English, and a search for what actually works. Tell us what you're trying to do.

Tell us your problem →