# Learning AI while working full time: what actually worked

At the start of 2026, AI was everywhere on my LinkedIn feed. Everyone seemed to be talking about it, building with it, or both. I was using AI almost every day, but only as a user. I'd type a question, get an answer, and move on. I never stopped to think about all the pieces being built and running in the background every time I asked something. Once I started wondering how it actually worked, I couldn't stop. So I decided to stop reading about it and start learning it properly, while keeping my full-time job.

This post is what I wish someone had told me before I started: what worked, what didn't, and what I'd do differently.

## Where I'm coming from

I'm not from a computer science background. I did my master's in electrical engineering at IUPUI and graduated in 2014. When I started working, I taught myself Java from scratch, mostly by building things and getting stuck a lot.

For the past seven years I've worked on infrastructure, where I had to learn something new all over again: large-scale systems, microservices, networking, and working with many teams at once. None of it came easily at first.

But I've always been drawn to the product side, to the question of what people actually need and why. AI finally gave me a way to act on that. I could take problems I face day to day and build something useful for them, start to finish.

So when I turned to AI, I wasn't starting with confidence. I was starting with something more useful: I knew what it felt like to be a beginner, and I knew that feeling passes.

## What didn't work first

Self-paced courses. I tried several, and none of them pushed me far. I'd start strong, then work and life would take over, and nothing was waiting for me if I skipped a week. I learned that motivation wasn't my problem. Structure was.

## What actually worked

**A cohort course with real deadlines.**
I joined a six-week cohort course with about seven hours of live online classes and an individual project due every week. Having a real deadline each week did what no self-paced course had done for me: it made me finish things.

**Building something real every week.**
Every week ended with a working project, so I never stayed in "just learning" mode for long. Two of them:

- **[Pack Your Bags](https://github.com/aamaninemtur0119/pack-your-bags)**, a trip planner where several AI agents work together. One checks visa rules by searching a set of official requirements and comparing its answer against live sources, so outdated information doesn't win. Another finds flights and hotels, and a third suggests restaurants and activities. It asks a human to approve before it adds anything to a calendar or sends an email.
- **[Dripstop](https://github.com/aamaninemtur0119/dripstop)**, which reads a bank statement, finds recurring subscriptions, and suggests which ones are worth cancelling.

**Learning the most from what broke.**
The demos always worked. Real use didn't, and that's where I learned the most:

- When I asked the AI to write links into its summary, it would sometimes reword them or drop them. I stopped relying on it for that and showed the links directly from the data instead.
- Someone could ask for a Japan visa while picking Toronto as the city, and the planner would happily produce nonsense. So I added a simple check up front that catches requests that don't make sense.
- The model sometimes returned its answer in a slightly broken format. Instead of letting the whole plan fail, I added a step that retries, repairs the answer, or falls back to a safe default.

None of these were in any lecture. They only showed up once I built something and used it like a real person would.

**Protecting fixed time.**
"When I have time" never worked for me. Setting aside fixed time for learning each week, and treating it like any other commitment, did.

**Leaning on what I already knew.**
Teaching myself Java and later learning large systems left me with habits that carried over: breaking big problems into small ones, reading errors carefully instead of guessing, and expecting things to fail so I plan for it. That last one turned out to matter a lot with AI.

**Learning alongside other people.**
Being in a cohort meant I could see how others approached the same problem. Seeing different solutions to the same project often taught me as much as building my own.

## What was hard

**Thinking like a product manager, not just an engineer.**
Every project started from a problem I actually had. Planning a trip meant juggling visa rules, flights, hotels and restaurants across a dozen tabs. Subscriptions kept quietly charging me for things I'd forgotten about. Building something that worked was one challenge. Deciding what would actually be useful was harder: what to include, what to leave out, and what a real person would want to see first. As an engineer, my instinct was to add more. As a product person, I had to keep asking whether it made the experience better or just more complicated.

**Getting the right data.**
The ideas were easy. Finding data I could actually use was not. Public datasets were often incomplete or out of date, and the data I really wanted wasn't freely available. A lot of my time went into finding, cleaning and working around data instead of building.

**Not being able to push things as far as I wanted.**
With a full-time job, I had to keep every project small enough to finish in a week. I could get something working, but taking it to the scale I had in mind, more users, more sources, more real-world testing, wasn't realistic in the time I had.

**Testing whether the AI was actually right.**
This was the hardest part. With normal code, a test passes or fails. With AI, the same question can get a slightly different answer each time, and deciding whether an answer is good enough takes real judgment. Setting up proper checks took more time than building some of the features, and running them was slow and sometimes unreliable. I learned that working in a demo and being trustworthy are very different things.

## If you're starting now

- Be consistent, and treat every piece of work as a chance to learn. Small, steady progress adds up faster than occasional big pushes.
- If self-paced courses keep stalling for you, try a cohort with deadlines.
- Build something every week, even if it's small and messy.
- Start from a problem you actually have. It keeps you honest about what's useful.
- Use what you build like a real user would. That's where the real lessons are.
- Expect data and testing to take longer than building. Plan for it from the start.
- Put learning on your calendar like a meeting you can't skip.

## What's next

If there's one thing this year taught me, it's that being consistent and learning from every bit of work you do matters more than anything else. Not every project went the way I planned, but each one taught me something the last one didn't.

I'm still early in this, and that's the point of this blog. I'll be writing about what I build and learn along the way, the parts that work and the parts that don't. If you're learning AI alongside a full-time job, I'd like to hear what's working for you.