About a month and a half ago, I unboxed a Mac Mini and had no idea what I was doing.
I'm not being humble — I literally had zero coding experience. My background was service industry: restaurants, hospitality, managing people, solving problems on the fly. I'd never written a line of Python. I'd never deployed anything. I didn't know what an API was beyond the acronym.
Today, I've got two live AI agent systems running. One's a trading agent that makes real decisions. The other is ShipStack with 5 active pipelines processing actual workflows. And I want to be completely honest: the first two weeks were me staring at my screen wondering if I'd made a huge mistake.
This is how it happened.
The Moment I Realized I Was Starting From Actual Zero
I bought the Mac Mini because I'd been reading about AI agents everywhere — the idea that you could give a system a goal and let it figure out how to accomplish it fascinated me. Coming from service work, where you're constantly making decisions and optimizing processes, it felt like something I could actually understand once I learned the mechanics.
First day: I opened Terminal. The cursor blinked. I had no idea what to type.
I Googled "how to start coding with no experience" and got 47 different answers. Python tutorials that assumed I knew what a variable was. Discord communities where people threw around terms like "async/await" and "middleware" like I should know them. GitHub repos with READMEs that started with "Obviously, you'll want to fork this and…" Obviously? I didn't even know what forking was.
My biggest fear: I'd wasted money on hardware and time on a dream that wasn't for me.
Week 1: The Overwhelming Part
I made a list of what I needed to learn:
- Python basics
- APIs and how to call them
- How AI models work
- How to structure an agent
- How to deploy it
That list made me panic. Five things, each one a rabbit hole.
So I did what I would've done in the service industry: I picked the one thing that would unblock everything else. For me, that was understanding how Claude's API worked. Not why it worked — just how to call it and get a response.
I found Anthropic's docs. They were actually clear — shockingly clear. There was a Python example that showed:
- Install the library
- Set your API key
- Call the API with a message
- Get a response
It took me about an hour to get my first successful API call. And when I ran that script and got back a real response? I got legitimately excited. That was real. That was something.
Wasted time that week: trying to understand machine learning theory. Watched a 2-hour YouTube video on neural networks. Useless for what I was building. Learned that the hard way.
Week 2: Building the Trading Agent (My First Real System)
Now that I could call Claude, I wanted to build something real. Something that would actually do something.
I decided on a trading agent. Simple premise: give Claude access to market data, let it analyze trends, make buy/sell recommendations. It felt ambitious, but in the service industry, you learn to commit to what you're doing even when you're not sure how.
This is where I got lucky: I found example repos of agents built with Claude. I didn't just copy them — I read through the code line by line and asked myself "why is this here?" whenever I didn't understand something.
Key things I learned:
1. Agents need tools. The agent's just a language model. You give it tools (functions) it can call. For my trading agent, that was get_market_data and analyze_trend.
2. You need a loop. The agent generates thoughts, decides what tool to use, uses it, gets results, and loops until it has an answer. This was mind-blowing. It's like a person thinking out loud and checking resources.
3. Error handling is 40% of the code. Half my trading agent code was "what if the API fails?" "What if the data's malformed?" I'd done this in restaurants — anticipating everything that could go wrong — but seeing it in code was a revelation.
I built the trading agent over days 5–7. Got it running live on day 8.
First real output from a live system: it recommended holding a position based on moving averages and momentum. I didn't follow the recommendation (too risky), but it worked. The system took raw instructions, fetched real market data, analyzed it, and gave me a reasoned decision.
That moment — watching a system I'd written from scratch actually produce meaningful output — I understood why people get obsessed with building.
Weeks 3–4: The ShipStack Breakthrough
High off the trading agent success, I wanted to build something more practical. Something I could use every day.
I'd been reading about workflow automation and realized: my service background made this obvious to me. I'd optimized restaurant workflows for years. Orders come in → assign prep tasks → track completion → close tickets. That's a pipeline.
So I built ShipStack: an agent system that manages order processing through multiple stages. Five pipelines:
- Order intake — Agent validates incoming orders
- Routing — Decides which fulfillment method (ship, pickup, hold)
- Inventory check — Confirms stock
- Assignment — Routes to appropriate team/system
- Tracking — Monitors completion and flags issues
Here's what's important: I didn't write all the code from scratch. By week 3, I understood the pattern. Create tools that do specific things. Create an agent system that chains tools together. Build feedback loops so the system can refine decisions.
I reused a ton from the trading agent. Different domain, same structure.
Took 3 days to get the five pipelines live. First day was integration headaches — connecting to actual data sources, handling different formats, dealing with real-world chaos. Day 2 was debugging. Day 3 was watching it run on real orders.
Once it started working: it handled 40 orders in the first test run with zero manual intervention. Assigned them correctly. Flagged inventory issues. Provided tracking updates.
I'd gone from "what's an API?" to "I built a system managing real business workflows" in a few weeks.
What Actually Worked
Learning by doing, not by studying. I didn't finish Python courses before building. I asked Claude for specific problems as they came up. This sounds like "fake it till you make it," but it's actually more efficient — context sticks.
Starting with existing patterns. I didn't invent the agent loop. It exists in docs, examples, tutorials. I read them, understood them, adapted them. Standing on giants instead of building from the ground up saved weeks.
Leaning on my existing skills. Service industry taught me to think in workflows, anticipate failure points, and optimize for reliability. That translated perfectly. I wasn't learning workflow thinking; I was learning how to express it in code.
Building incrementally. The trading agent was Version 0.1. ShipStack started as just order intake, then I added pipelines. Each piece worked before adding the next.
What I'd Do Differently
Skip the theory initially. Neural networks, transformer architecture, the math behind LLMs — interesting later. Useless day 1. I wasted 4 hours learning things that didn't help me build.
Read docs over tutorials. YouTube tutorials are fun but outdated fast. Docs are the source of truth. Felt harder at first; was actually faster.
Ask for help earlier. I struggled alone for 3 days before posting in Discord. Got answers in 2 hours. Pride's expensive.
Start with something you actually care about. I built a trading agent because markets fascinated me, and ShipStack because I wanted to automate my own business workflows. If I'd picked a random project just to "learn," I'd have quit week 2.
The Weird Part
The strangest thing about this journey: I still don't feel like a "real" programmer. I've never taken a CS degree. I can't explain what a hash table does. I've never optimized a database query.
But I've built two live systems that handle real workflows. They work. They scale to the load I'm giving them. People could use them.
I think that's the thing nobody tells you: you don't need to be a "real" programmer to build real things. You need to understand the specific domain you're working in, care about the problem, and be willing to get uncomfortable.
I had all three from my service background. The coding was just the syntax.
What's Next
I'm building memory into ShipStack — giving it the ability to remember things across sessions. Right now every conversation starts from zero. The agent doesn't know what we talked about yesterday or what it's written before. That changes next week.
If you're sitting where I was seven weeks ago — service background, no coding experience, fascinated by agents but terrified to start — here's what I'd tell you:
You're not starting from zero. You're starting with every problem-solving skill you've built in your actual life. You just need to learn the syntax to express it. That's not four-year-degree hard. It's seven-week hard.
Buy the Mac Mini. Open Terminal. Open Claude. Ask one question at a time. Build something real.
It's the best investment I've made.