So, I've been using all sorts of coding agents lately and have some conclusions. Worth reading if you're into this

CLI Agents (the true way, apparently)

✳️ Claude Code - still your primary workhorse, the goat. Expensive, sometimes nerfed and buggy, but the best out there. Especially if you can use Opus. Best in following multiple priorities and reqs. Can integrate in IDEs, but I disabled it. Sub worth every buck

🤖 Codex - new player. Worst CLI UI/UX, worst functionality (no subagents, no custom cmds?). But the sanest model after claude - gpt5. Now does about 10-15% of the job for me. Highly inconvenient limits. Once reached and you locked out for many days! Couldn't commit properly (\n\n everywhere), fixed with prompt lmao

🐙 OpenCode - the rising OSS king. CAN use your Claude and/or GitHub Copilot subs! I repeat: just use Claude Code sub in it! Or gpt5 from Copilot sub! Supports ANY model! Compatible with Claude subagents and commands. BEST subagents functionality atm. I'm starting to use it more and more. However, current access control is either yolo or spam for every bash call. Couldn't set it up conveniently. What am I doing wrong? UX is questionable, but unique and sometimes really thoughtful. Plus it's OSS!

✨Gemini - your super fast crazy giant. Amazing speed, huge context window. But can nuke your codebase in a blink of an eye. Or could start to whinу and ask to kill it 😵‍💫 I use it for 2nd opinion for reviews, and never give it rights to modify anything lol. At least it's free (if u ok to give them ur data for training 😂)

🐲 Qwen - fork of Gemini CLI, but with qwen under the hood. Also, another rising workhorse for me. Solves simple to medium things reliably. Gives nice distinct view in reviews. Testing it in more complex scenarios. Has pretty generous free limits (same concern for data privacy as with google)

IDE Agents

💎 Cursor - top choice for the Q1. Not anymore, though - I cancelled sub. Still the best autocomplete out there, Cursor Tab is pure magic! But who writes code manually now? 😂 Bugbot from Cursor was the best reviewer for me, but pricing model is not suitable. Not fair, I know..

💡Junie and AI assistant in IDEA products - finally, they do their job (after 2 years of uselessness), but nothing really special. Gunie, AI Writing Companion browser extension is in the same subscription and does the best job in browser imo. It's the only real reason to use it (for me), lol

Magical prompting still does its job, but hacks changed a lot from last year, for reasoning models. You can still do triple more if prompting properly, use ADR/SOP, etc. Context management becomes more and more important

My next tries - Conductor, Claude in Zed, more subagentic workflows. What else should I try?

your feelings? wdyt?

@art_of_dev

Gave two talks on droidcon Lisbon a week ago. Both were on topics related to Android development. Nice conference, productive networking. Unfortunately, there are no records

Ping me if you're interested in the presentation. The material for the 2nd one turned out to be especially interesting and insightful

Maybe in Berlin next time? We'll see

One of real trouble why Ru-founders suck is they don’t believe in capitalism.

If I pay my friend startup $100, and he will pay $100 to my startup, we will both have revenue, and traction.

Repeat that with 10 friends and you got $1000 MRR

Repeat that with 100 friends and it is $10,000 MRR

How much can you raise if you have $10,000 MRR? some.

But the issue many folks want to get all for free, discounted, and don’t even pay respect if someone offers something for free just because you like them

Wanna try some new agentic browsing experience?

Got access to Comet — AI browser from Perplexity. It is invite-only, so I'm sharing with you.

Here is a first invitation:
https://perplexity.ai/browser/claim/PK37V8K37R

dm me if it's already taken and you need more (I have 5 invites total)

ngl, I'm still staying on abandoned Arc, but it's nice and insightful to play with Comet. If your main browser is a usual Chrome — it can be immediately better and more powerful!

@art_of_dev

So, if you haven't seen it yet, OpenAI has released Sora 2, its latest video generation model. It directly competes with Veo3 thanks to its sound generation capabilities and decent physics and world understanding. Overall, it's a very interesting model to experiment with.


My two cents:

• It's an incredibly powerful tool for indie hackers, builders, and founders of all kinds.

• Absolute wild powerhouse (similar to Veo3!) for testing ads, marketing, and any potentially viral ideas for reels, stories, or anything creative.

• If you are building smth, or looking to marketing — definitely experiment with it and see what others are doing.

• It's crazy, absolute AI slop garbage and mostly, mind-melting brain rot 😂

• But you can learn hell a lot. Focus on possibilities, not on melting your brain! Stay safe, bros and gals 💪😁

• Acrobatics and other complex motions are still the ultimate test. Every current video generation model is bad at this, including Sora2 and Veo3. But most people can't tell the difference already.

• There is NO official API for Sora2 yet! All integrations are unofficial hacks, at this moment, so be careful.

• Every generated video has animated visible + invisible Sora watermarks

• Sora has a TikTok-alike iOS social app — only accessible with an American or Canadian apple account.


✨INVITE to SORA 2✨

Official usage is INVITE ONLY!
It's becoming a nice tradition - I want to share with you.
There are 2 slots available with my code: => KN0W5B <=
If you use it, pls share your own invite code in the comments below to include more people!

How to use:
1. Turn on your American VPN
2. Go here: https://sora.chatgpt.com
3. Log in with your Free/Plus/Pro account
4. Enter working invite code
5. Generate anything! ✨

More invites:
You can find more invites on Telegram (ru!) where people share their invite codes.
Rules are simple:
1. take one code
2. get access
3. share your own invite code below in comments AND in the thread where you took it
4. PROFIT, everyone is happy 🕺

Invite sharing threads:
- https://t.me/seeallochnaya/2964
- https://t.me/voicestuff/1082
- https://t.me/mobile_ads_ai/1054

P.S. I'll post more generated examples in the comments.

@art_of_dev

Ok, Sora 2 is officially in the API ($0.1-$0.5/sec)

GPT-5-Pro is too, but it's too expensive to be used on average anyway ¯\_(ツ)_/¯

But what's really huge is the new OpenAI Apps SDK. Now you can embed your own web app in ChatGPT. It understands every piece of context and every user action, allowing for seamless, smart usage, whether textual or voice-controlled.

It's sort of a much more powerful version of custom GPTs. Chat will now automatically recommend the most relevant apps. This opens up ChatGPT as a real, full and powerful second-layer app market. It's already integrated with payments to some extent, so I think there'll be a revenue-sharing with OpenAI. But it's a new market and a new source of traffic for us all. A brand new platform has just been born.

Live was here
https://www.youtube.com/live/hS1YqcewH0c?si=zzkFyo5KzqPKb7pU&t=1237

@art_of_dev

Sometimes, you can feel so down, lost and powerless, with no energy or focus. But reviews like this always uplift me and help me to move forward, even if it's just one small step at a time ☺️

Building something and feeling that people really appreciate it is one of the most wonderful feelings I've ever experienced!

@art_of_dev

If there's one thing I've learned from creating various products, it's that any builder needs to put their ego aside and accept the fact that they won't know how their products will be used or for what purpose.

The only way to actually find that out is to:
1. release something
2. talk to users and clients
3. analyse how they actually use your thing, what actions they perform and which problems they are trying to solve.

Indie hacking is a way of learning about the world and the people in it. It's a kind of real social science. You set experiments in the wild and study.

I try to remind myself of this constantly:

You know nothing. Be humble.

This is particularly true of programmers. As we are often detached from ordinary people so much, we think, choose and pay in completely different ways. We are an entirely irrelevant audience for almost any product. We are some of the worst possible users and customers to target.

@art_of_dev

And an alternative opinion for the sake of balance 😂

Wow, this is just beautiful

Super Heavy B15-2 start slow-mo

@art_of_dev

yc was right.

do things that do not scale , just following this rule:

- we did our first $2000 in sales without any product in a first week.
- we spent all money on R&D
- we had 6 clients from high-volume agencies, mostly CEOs or CMOs, with >$1M/mo spent.
such people are so valuable, so we learnt a lot.

we ship them demoes using python scripts, zip archives, and google drive.

some clients from waitlist refer us to other high-volume agencies, even before our intro call

each client showed us biggest gaps in the ads market.
everyone can generate AI videos, but only few knows how to cook them right.
we saw patterns, that will lead us to a real product

Modern problems require modern solutions!

Today, I saw the new format TOON (Token-Oriented Object Notation) that has been optimized for token efficiency. It could mark the beginning of the end for JSON as the go-to format for LLM-related tasks. Or not 😁

TOON formatting is deterministic, minimal and token-efficient. Typically 30–60% fewer tokens than JSON!

Here is an example of the same data in JSON and TOOM for comparison:

{
  "users": [
    { "id": 1, "name": "Alice", "role": "admin" },
    { "id": 2, "name": "Bob", "role": "user" }
  ]
}

users[2]{id,name,role}:
  1,Alice,admin
  2,Bob,user

I've checked with a few top models, and they were all able to understand this new format without needing any additional instructions. So it's ready to use right away, at least for reading.

What do you think?

TOON repo: https://github.com/johannschopplich/toon

@art_of_dev

Slowly learning the social marketing game.

My first post on threads that got 55K views, a hundred replies and a few hundred likes.  My max on Twitter/X is 1K views so far 😁

Btw, let's connect/sub if you use threads!

@art_of_dev