Hello! 👋

Today we have a well-known indie developer (since 2024), online course creator, one of the biggest newsletter owners, and a person with a huge social media following, Antoine van der Lee .

Antoine will share his tips on how to adopt AI as an indie developer.

Let’s dive in!

Follow him on X/Twitter & LinkedIn and read his Newsletter.

Create App Store Screenshots in Minutes with AppScreens

Indie developers should not spend days designing screenshots.

AppScreens gives you hundreds of ready-to-use templates, automatic resizing for every device, and one-click uploads to the App Store and Google Play.

Create a store page that looks professional and converts, without hiring a designer or an agency.

Design screenshots now!

AI Adoption as an Indie Developer

Where companies have a hard time realizing they don’t need 20 engineers anymore because AI takes over, I’m personally realizing I just hired 20 engineers as an indie developer.

The root cause? Agentic engineering.

A lot has changed and my journey went from adding shouldRasterize = true after a full day of research, to waiting for Swift 1.1 to complete building as there were no incremental builds. Today, I’m in the middle of a new era: developing apps using AI.

Reinventing ourselves

It feels a bit awkward sometimes. I used to be good at prioritizing. Especially when I worked full time at WeTransfer while I was building up SwiftLee (avanderlee.com) on the side. Side projects have limited time, so prioritizing is critical. When I went Indie in 2024, it suddenly felt like I had time for everything. I went from side projects to full-time, 5-day week focused on just personal projects.

The result was a big issue: I lost my planning skills, I did not prioritize, and I basically went on building all kinds of things at the same time. Note: this is pre-AI, at least, I did not fully embrace AI yet.

After a few months, I started to realize planning and priorities were still important. I invented Indie Antoine 2.0 and became the most productive version of myself. I launched my first course (going-indie.com), which, ironically, is full of tips on planning and prioritizing. At the same time, I launched a new venture for RocketSim for Teams and released several major versions. In other words, I found back my best productivity version.

Last December, I felt kind of lost. I had meetings with an Growth Accelerator coach and regularly checked in on progress. At a certain point she asked me “What do you want to focus on in 2026?” and I did not know. I was always good at planning, but I could simply not answer.

My blog visitors count is decreasing. Developers are embracing AI, so no need to read those articles manually anymore. My newsletter, at the same time, is growing faster than ever before. RocketSim is here to stay, but how will developers build apps in 2026 now that AI is moving so fast?

At the same time, I was embracing AI more and more. I started developing more with Cursor than with Xcode, and I wrote fewer and fewer lines of code. My CHANGELOG.md got more additions than ever and it felt like I could build anything that I want!

Do you recognize a pattern?

The same problem occured as when I went Indie in 2024.

AI opens so many doors, we can suddenly build almost everything.

Building multiple apps at the same time

And that’s where the trouble started. I worked on RocketSim, new course lessons, and a new app (Vydio) at the same time. I started working on my YouTube channel growth and felt untouchable.

Little did I know…

Context switching costs a lot of energy. I remember having days at WeTransfer where I worked 2 hours solely on coding, writing actual lines of code. It felt great, fulfilling, productive, but also tiring.

Now with Agentic coding, it feels like I have this every single day. At a certain point, this was my workflow:

1. Write a planning prompt in Cursor window 1

2. Wait for that plan to complete

3. Review another finished plan in Cursor window 2

4. Reply with review feedback

5. Go back to Cursor window 1 and execute the plan

6. Go back to Cursor window 2 and start a new prompt

Whenever I was done reviewing, the other window completed work. I was highly productive, but I also lost energy like crazy.

Reduce context switching, remain productive

What I realized is that it’s okay to work on many projects at once, but that there’s a difference in what you can work on at the same time. Building a new app while working on an existing one did not work out for me. I really have to deliberately pick the different projects that I work on.

Ideally, you work on the same project, but different features. This reduce context switching in the sense that you remain active in the same domain. Easily doable for iOS projects, a bit harder for a macOS app like RocketSim.

Getting better at Agentic Coding

I know you all want to learn how I use AI. It’s funny, I write an article every week, but I did not write a lot on agentic coding until recently. The reason is that I didn’t feel like I was knowledgable enough to share learnings yet. I felt like I had to wait a little more, as I was still learning a lot.

The reality?

I never stop learning and things are moving just blazingly fast. We recently had a week in which OpenAI and Anthropic released their new models, while we also received the first Agentic Coding support version of Xcode.

So I started sharing learnings along the way. The first SwiftLee articles already existing under a new AI Development category and I released four different Agent Skills for iOS development.

While many received those Agent Skills as welcome additions to their workflows, it has a different meaning for myself: I’m building my personal fundamental AI architecture.

I want to create an AI version of Antoine.

If I can teach AI to write code like I do myself, I can multiply myself, and develop faster. I’m constantly detecting repetitive tasks in my workflow, and finding ways to automate things. This is really a learning curve, a skill to develop.

The most recent example is my product analysis workflow. I normaly go into Amplitude, into RevenueCat, manually evaluating tracking events and sales performance. I’ve now configured their MCP servers in Cursor and asked:

Analyze the current codebase. Match the tracking events with Amplitude, get conversion metrics from RevenueCat, and analyze my sales view using the best-practice resources that I stored in /docs/references. Create a plan with high-impact improvements to the paywall.

This is not the exact prompt (I added a few specific file references, for example), but you get the idea.

I gave AI eyes.

That’s been crucial in my workflow improvements. The more context I could give to my agents, the better the outcome.

The next version of this is to add a recurring automation that checks in on these statistics, applies code changes, and opens a PR for a new experiment. Ideally, based on actual experiment outcome data (I’m working on this).

My workflow best practices

Apart from the journey I’m currently in as described above, I also have a few best practices that I like to apply:

• Always use plan mode. Even if I think “I can one-shot prompt this”, I’m still surprised by the outcome of a plan that’s often a little off on what I wanted to do. Plan mode prevents mistakes, reduces review time.

• My preferred model is currently Codex 5.3, but you’ll have to decide for yourself. I use it for both planning and execution and I’ve been happy with it ever since 5.2

• Give your agent eyes. Yes, stressing that by repeating it here. Whenever I ask an agent to fix a bug, I ask it to first reproduce it with a unit test.

• That brings me to: Test! Yes, write tests. Make sure tests are validated on success after every run. This speeds up validation as you know nothing broke with new changes.

• Configure linters and formatters. The last thing you want is to spend time on asking AI to apply code styling.

• Apply learnings. Constantly evaluate where you’re losing time. Ask the agent to “Update the AGENTS.md file based on the learnings from this task” so the same mistakes don’t happen again. Find out what costs you the most time during code review and see if you can reduce that time spend. These optimizations will compound over time.

• Use external reviews. With this I mean either adding GitHub Copilot as a PR reviewer or opening a new agent task to “Review the current unstaged changes using these Agent Skills”.

For the latter, I’m currently using my personal skills for SwiftUI, Swift Concurrency, Core Data, and Swift Testing. All these skills are based on my code projects, articles, and course lessons. The review prompt reference these skills makes sure that the local changes are close to what I would write myself. Here’s an example:

Whenever I find improvements I see manually missing, I’ll update the skills to prevent that from happening again.

Altogether, this will help my agents improve by 1% every day, making me much more productive in a year.

A new course: Agentic Coding Fundamentals for Developers

You can tell from this story that I’m on a journey and if you know me, you know I like to share my learnings along the way. That’s why I’d love to introduce you to my new course: Agentic Coding Fundamentals for Developers

The course is in active development, and you can join the waitlist at aifundamentals.dev. While AI moves fast, I’m sure that new modules and lessons will appear. However, I want you to have a way to stay up to date without having to go through the learning curves I’m having. We should all become more effective at agentic coding and I can’t wait to share you all the best practices.

Want to learn more? 👇

Follow him on X/Twitter & LinkedIn and read his Newsletter.