I've spent a lot of time trying out different AI models for different tasks. Here's a list:
1) Claude 4.5 Opus thinking - To build a project from scratch; great to scaffold.
2) GPT 5.2 low reasoning - For minor fixes or to add new features according to the PRD.
3) Gemini 3 Flash high reasoning - For designing the UI/UX from scratch.
4) Gemini 3 Flash low reasoning - To improve the UI elements.
5) GLM 4.7 - For documentation.
We are revising our developer API policies:
We will no longer allow apps that reward users for posting on X (aka “infofi”). This has led to a tremendous amount of AI slop & reply spam on the platform.
We have revoked API access from these apps, so your X experience should start improving soon (once the bots realize they’re not getting paid anymore).
If your developer account was terminated, please reach out and we will assist in transitioning your business to Threads and Bluesky.
The wait is over ☀️
Brightside is now LIVE on the App Store.
Trade everything anytime, anywhere, right from your phone.
Download on iOS and get started now ✨👇
@Jpandya26 The abundance of short-term grants/prizes has really ruined Web3. Students are happy to chase 2-3K USD with no long-term vision. The whole space seems like a farmable incentive loop with students optimizing for quick cash, and nothing built survives past demo day
This is a solid idea, even I was thinking about it but let's say if too much capital moves to the vault, the CLOB becomes thin → spreads widen → slippage increases → the market quality drops. Don't you think the CLOB would actively need orders specially during high volume events
A few months ago the Indian government asked WhatsApp to break encryption and provide a backdoor for them. WhatsApp refused, citing privacy concerns. Government propped up a new chat app which now has 30M+ downloads.
Now they’re trying to force an app on everyone’s phone. Again, Apple has refused to comply citing privacy concerns. Now journalists and IT cells are running smear campaigns on “foreign apps.”
It’s abysmal that private companies care more about our people than the government and yet the people instead of revolting are defending these actions. Wake the fuck up, protect yourself.
Who is building the x402 facilitator using @KRNL_xyz
Why you should use KRNL,
-> minimize trust assumption with cryptographic proofs, which will be verified before settlement
-> cross-chain support out of the box, you don't need a separate config/provider setup for each chain
-> KRNL uses EIP-4337 to settle the transaction, so you can also let the user/ai agent pay for the gas directly
I will be spending some time on this and will build a oss poc out
I agree to this, it might be useful to use TOON to represent simple data structure, but for complex data, TOON is a complete disaster.
It would be very hard for LLMs to understand the relationship between data pointer which will lead to hallucinations,
I think llm tokens will be commoditized in the near future as there would be hardly be any difference between the LLMs out there now and at the end of the day they will start a price war to capture the market share
TOON is doomed to fail.
TOON struggles with nested data. Independent tests already show accuracy dropping compared to JSON. The moment your structure gets complex, TOON becomes a liability.
On top of that, JSON is the default structured language. That’s the native dialect.
JSON appears constantly across the internet
>APIs return JSON.
>Dev tools save logs in JSON.
>Configs, datasets, telemetry, and modern documentation rely on JSON.
>GitHub is full of JSON snippets and structured examples.
So models see JSON thousands, even millions, of times during pretraining. So, you don’t get bonus points for switching to TOON. In some cases, you get worse results.
Even if TOON worked perfectly, model providers could kill its moat overnight by compressing JSON themselves, adding new wire formats, or absorbing TOON-style ideas into their protocol layer.
Then there’s human reality. JSON survived because anyone can skim it at 3 a.m. TOON needs new tools, new habits, and attention nobody has time for.
At the end of the day, TOON is more of a middleware trick.
The real winners will be systems that handle human chaos and optimize behind the scenes, without asking anyone to change how they speak.
KRNL Studio plays a huge role in abstracting the concept of DSL and helps you focus on the core business logic of your dApp.
I will be making more such explainers to make the life of the devs building on KRNL easy!
Building with KRNL: Intro to KRNL Studio 🎥
In this walkthrough, our DevRel engineer @ash20pk breaks down how developers can visually design complex, cross-chain workflows without touching JSON.
KRNL Studio lets you:
• Build KRNL DSLs with drag-and-drop steps
• Connect Web2 APIs + Web3 functions in one workflow
• Auto-generate contract calls and mappings
• Export/import workflows locally, no data ever leaves your browser.
This is how we make verifiable orchestration simple, fast, and developer-friendly.
Watch the full demo below 👇