Agentic coding for mobile developers simplified - you're not as far behind as you think
· Agentic coding demystified
I know, I know, another blog post about AI and agentic coding to clutter your feed and swarm you with information you don’t feel like you need. You’ve been hearing about AI, you must have, but you either tried it back when GPT-3 was a thing and it would make silly mistakes all the time, or you’re constantly being bombarded with people on Twitter telling you that AI is cooking their brains, or telling you that they’ve created a 10k MRR SaaS B2B app and you’re a sucker who doesn’t use it yet.
I know, you’re tired of it, but it feels inevitable. So, you want to at least get a little bit of knowledge about it, but don’t know where to start. Well, that’s what I’m here for. Let’s take a bit of a step back from all the noise and the faff, and distill the situation with AI agentic coding to the basics, stuff that you can use right now, and what you should care about.
Disclaimer: This is a very, very fast-moving industry, with very dynamic changes. The content of this blog post is accurate at the time of writing, and I will attempt to edit it as things change, but bear that in mind.
What is agentic coding? Is it ChatGPT in my editor?
Let’s cover the basics first, to make sure we’re all up to speed on terminology and the general vibe (pun intended). Agentic coding is, spoiler alert, coding with an agent. The main difference between what you might be used to with just ChatGPT is that an agent can have reasoning levels, and can complete a full task end to end, prompting itself, and basically “take control” - write to multiple files, delete files, create new ones, run tests, run terminal commands, etc. It can, but it doesn’t have to. You control how much power it has. There’s of course more to it than that, it’s an iceberg, but for our use case, this is the gist that you should care about.
Where am I using it?
This can be in a few places, but to make things simple, let’s simmer it down to 3 locations:
The IDE
The current IDE you’re working with has an agent mode connected to it. It’s not a “probably it does”, it absolutely already does. Android Studio has Junie / AI Agent / Claude integration (we’ll discuss it later), and Xcode recently received the ability to have agentic coding in it as well. Hell, there are even tools like Cursor and Antigravity that are entire IDEs dedicated to AI coding. Wherever you already are, there’s agentic coding involved in it.
The terminal
This is where a lot of these tools started, and where a lot of them still live. Each one of the “big players” has their own CLI too, and there are tools like OpenCode that can be a central point for all of them. However, to keep things simple, we will be talking today about the first-party CLI tools, the ones the companies created for themselves.
The standalone apps
Again, the big players here have their own standalone apps. These apps are, in reality, a wrapper around their CLIs, but they make installation a lot easier, provide a nice GUI, and make project management easier.
Who is them?
Right, yeah, time to talk about it. What actual agents are we talking about here?
Well, there are honestly a bucket load, and it’s partially why this sphere is so daunting for so many people. In reality, there are only two players you should care about: OpenAI and Anthropic.
Are there other providers? Yes, there are.
Are they close in performance? Some.
Do you need to follow all of them? Absolutely not.
In reality, these two are the biggest players out there, and for good reason. Their models (Codex 5.3 for OpenAI, and Sonnet and Opus 4.6 for Anthropic) are highly optimized, really accurate, and honestly do the job. There are other good providers, like Google for example, but I’m going to focus on these two because they’re the gold standard.
Where should you be using them?
It does not matter. Surprising, right?
Well, each of these platforms has its pluses and minuses, pros and cons, and you might find that you prefer to, for example, stay within your IDE for all of your development. That’s fine. You might find that you like the simplicity of the terminal. That’s also fine. Give them each a whirl, take a day to play around, find the flow that works for you, and that one is the best. Can you optimize things? Probably. Endlessly. You can spend all day optimizing the flow for best performance and execution and whatever. Or you can, and in my opinion should, fuck around and find out.
Which one do I choose?
Ready for this? It does not matter.
I know this might be a hot take for some, I know the die-hards for either platform are fuming right now and are writing me angry emails (or using their provider of choice to write me an email), but in reality, for your day-to-day usage, it really does boil down to which one you like more. Once more, this is a case of fucking around and finding out.
“But so-and-so on Twitter said that if I don’t use Opus I’m a pleb.” It doesn’t matter.
“But they told me on LinkedIn (what?) that Opus is slopus!” It doesn’t matter.
“But this dude said that…” IT DOES NOT MATTER.
For 99% of your time, you will not feel any difference between them. Try them both, they each come with a trial period, see which one jives with you more, and go with that one.
Takeaways
I think it’s quite clear: fuck around and find out.
Install some agents, in your favorite workspace of choice, or in a new one like the terminal, and see how you go.
In the next few weeks, we’re going to go down the path of agentic coding. I’m going to hold your hand and tell you that you’re doing great, and by the end of it, you’ll have enough knowledge to feel confident being a programmer in today’s day and age. Or enough knowledge to know you prefer to move to a farm in the woods and leave this whole thing behind. That’s up to you.