For the last month, I've been trying out AI tools for software development in my spare time, mostly on evenings and weekends. I built an app, a game (also an app), a few websites, and a bigger financial software project. I did all of it without writing a single line of code.

Some projects went well. Some did not. Some were easy. Some were hard. Here is what I learned.

The first feeling: anxiety

AI makes you move fast. Really fast. Things that used to take days or weeks can show up in minutes.

That speed comes with a new kind of anxiety. You are not always sure the product is solid. You are not always sure how the system works end to end. You are not always sure about security. The app works, but you keep wondering if it works for the right reasons, or if you just got lucky.

This feeling might be stronger if you built software before these tools existed. You know what "normal" looks like, so you can feel when you are skipping steps.

The second feeling: "I can build this"

Over the years I built a big backlog of ideas in my head. The main problem was always time.

A decent side project can easily take 100 hours. Now, that same scope can sometimes take 10 hours. Not always, but often enough that it changes your mindset. Ideas that felt "nice but unrealistic" start to feel possible.

System design still matters a lot

AI can build quickly, but it can also create chaos quickly.

You can end up with a messy architecture in one afternoon. You can accidentally expose your OpenAI keys in the frontend. You can ship something that "works" but is fragile.

You still need to lead the design. You still need to define boundaries, data flow, and deployment. The agent can help, but it will not protect you from bad structure unless you tell it what good structure looks like.

Tasks might be over, but jobs are not

In every project I built, nothing happened automatically without me. I was the product manager. I was the designer. I was the engineer making decisions. The agent was a force multiplier. It was there to melt tasks, not to replace ownership.

The work did not disappear. It shifted.

What is easy now, what is still hard

1) Personal website and blogging: very easy

A personal website for branding and blogging is almost trivial now. WordPress already feels outdated for this use case.

Buy a domain, rent a VPS, and talk to your agent for 30 minutes. You can have a clean site running fast. Agents can SSH into your machine, set up Nginx, and configure HTTPS certificates.

Still be careful. If your site is not fully static, you should consider a security review from another agent.

Time spent: about 1 hour.

2) Mac utility for Alt-Tab and window switching: very easy

I built a small Mac utility that makes Alt-Tab and window switching easier and faster.

Time spent: about 1 hour.

3) "Agent Assist" visualizer: very easy

I built a small tool that visualizes what your agent is building, so you can see progress and context while it works.

Time spent: about 1 hour.

4) iOS app to track toddler vocabulary: easy

Pediatricians often ask how many words your toddler speaks. That question is harder than it sounds. Most parents do not have a good estimate. Is it 50? Is it 300?

I built an iOS app to track it. About 30 views so far. It has premium features. All data is local. No servers, no cloud.

An app with this scope can now be built by a PM or designer in a few days, if they are comfortable guiding agents.

Time spent: about 8 hours.

5) Roguelike text-based bureaucracy iOS game: medium

Games are different.

Agents can write design docs and implement systems. But they do not know what is fun. They do not feel pacing. They do not have taste.

This was my most enjoyable project because I was involved constantly. I was deciding every minute. I was the designer. The agent executed what I asked.

Time spent: about 16 hours.

6) Web product with real users: medium

This gets harder fast.

You need to understand how a web app works. Where data lives. How auth works. How models relate. What should be in the client vs server. How you handle edge cases.

Yes, you can ask agents to design architecture. But you still need enough understanding to judge whether the design makes sense.

If you already know some basics, almost any consumer web app MVP can be built in a couple of weeks.

7) Algorithmic market maker, backtester, paper trader: hard

I had this idea for a long time. I tried earlier AI tools and failed. I tried again recently with the "no code" approach. It is not finished, and I have doubts.

This is a complex system. It pulls price and order book data from multiple exchanges. It needs clean storage, correct algorithms, backtesting, and paper trading. The scale is large, and mistakes are dangerous.

I used review agents to catch issues, but the surface area is huge. For this kind of project, I still think I would need to write parts of the code myself, or at least go much deeper than prompts.

AI is close, but it is not fully there yet for large, high-risk systems.

Final thoughts

We will still be here. We will still be building.

The gap between average and excellent builders will grow. Juniors will have a harder time, not because the work disappears, but because the bar moves up. Code will become cheaper. Product thinking and system design will not.