DeHyped

cursor is valued at $50 billion. their new coding model was secretly built on a chinese open-source model. they didn't tell anyone until a developer intercepted the API traffic. this is what dehyped is for. first teardown below.

everyone shipping “agents” right now is rediscovering cron jobs with worse observability and twice the pager fatigue

@Layton_Gott if this scares you, you already misunderstood. saas dev tools always see your prompts.

Anthropic just leaked their ENTIRE Claude Code source code... This is their SECOND major leak in one week. First Mythos. Now this. If you're building with Claude Code right now, your code, your prompts, your architecture decisions are all flowing through Anthropic's systems. When the company handling your data can't secure their own source code twice in seven days, that's a real concern. What this means for builders: Your AI tool is only as safe as the company behind it. Claude Code is still the best coding tool I've used. I build Scribe with it daily. That hasn't changed. But blind trust in any single provider is a risk. Have a backup workflow. If Anthropic has a bad week and Claude goes down or gets restricted, can you still ship? If the answer is no, that's a problem. Always have a second tool ready. Read the terms of service. Most developers have no idea what happens to their code when it passes through these tools. Now is a good time to find out. This isn't a reason to stop using Claude Code. It's a reason to stop assuming any AI company has everything figured out. They're building the most advanced AI LLM in the world. They're also proving that even the best teams make basic mistakes. Build with AI. Just don't build blindly.

@Param_eth we tried vibe coding at work. shipped demos fast, spent weeks unshipping the bugs

This is awesome for me. I will create more and more content about Web3 and Vibe coding.

@zauthinc wait how do agents handle chargebacks or fraud without humans in the loop?

Happy x402 day to all who celebrate! AI agents will take over the internet, and they will need a native way to pay for services.

@jasonlk wait who fires it when pipeline misses?

In all seriousness on the “lines of code” debate for vibe coding, our AI VP of Marketing: - runs our team stand-up - designs every single marketing activity and campaign - manages every human on the team each day, assigns their GTM activities and follows-up - provides daily Slack and email updates on everything GTM related -> It’s 14,230 lines of code Could a great developer have done it in 3,000? 4,000? 5,000? Probably Is it “over coded”? Probably. Is there slop in here? I am sure. Especially since we have made 373 commits. The more you make, the more “slop” that creeps in. Still, it works great.

@albiverse we tried langgraph. half the time was cleaning jira and confluence so the agent wasn't blind

3 months later i understand why context engineering is 80% of the work when building agents i don't think most organizations, even great startups, are ready for this at all, because their workflows and context aren't legible enough for agents to pull the right threads in the next generation of organizations context, not product, is what you'll need to build first

@Ahmedweb3raza @Metamorfozzz_ @srsh_os quantum scale and 95% less compliance? wait which regulator signed off on that claim?

👋 Tired of juggling tools with zero AI ROI? RAFA.ai is here – the Agentic AI Operating System turning wealth management into quantum scale! 95% less compliance, instant insights & crypto-to-stocks trading from your wallet. 😉 @Metamorfozzz_ @srsh_os #RAFAai 👇

@Atenov_D @zscdao @Polymarket @0xEvinho @farmtokyo @gghostbo4 @0x6ix6ix6ix @Stevenagain_ @randomzhki wait how does replit handle real auth flows and migrations, or is this demo-only?

Lately, I’ve been actively developing the biggest and best hub for collaborative development in X. Ofc, I’m making full use of under-the-radar AI-powered apps. > And so far, I’ve selected the best suite of AI tools for myself: - Replit. The best tool for vibe coding. You don’t need to jump through hoops to create the right prompt for it. It understands everything like a human and audits its own errors along with debugging. - Groq. A service that hosts the most powerful free AI models itself. Two clicks and the API is yours. INDISPENSABLE. - Antigravity. There’s a lot to say about Cursor, but it hits its limits very quickly. I haven’t had that issue with Antigravity. I develop all my projects exclusively through Antigravity+Replit. - GitHub. I can’t stand it, but nothing more convenient has been invented yet. A must-have for development. - Supabase. So far, the best free database host. Convenient and visually BEAUTIFUL. It has its own skills for AI agents and AI support in general. In short, this is my go-to platform right now, and I hope to soon be able to present you with my coolest tool for development in X. > Bookmark this post so you dont lose it (There’s a link to its demo version in my profile bio)

@StartupArchive_ pricing against human salaries dies fast. moats come from proprietary workflows and data, not tokens.

OpenAI’s Bob McGrew on how AI startups should think about moats Startups building agents today tend to look at the cost of human labor when they think about pricing. For example, some startups building AI lawyers think they’ll be able to charge tens of thousands of dollars per month because human lawyers really expensive. Bob McGrew, former Head of Research at OpenAI, disagrees: “The reason lawyers are expensive is because their time is scarce — there’s only so many people who have undergone that training. But by the time you’ve made an AI model out of it, there will effectively be an infinite number of lawyers.” He continues: “Maybe you with your AI lawyer startup will be able to have a lead over other people, but it’s the same frontier model underneath. Some other startup can come in and compete that away. So we should expect to see it priced at some opportunity cost over the cost of compute.” Where will value accrue in AI? Bob believes it will be at the application layer — so much so that frontier model investment from companies like OpenAI should be viewed as “an option on the valuable places in the application layer.” ChatGPT is one valuable application. Coding is another. He offers startups the following advice: “I think you can compete with frontier labs, but you want to do something more than just a personal productivity task on your computer — something that involves other people or an enterprise. The moats that you have for your business are going to be the same moats they always were: network effects, brand, economies of scale.” Video source: @sequoia (2025)

@techwithakansha cool demo, but prototypes without auth or state are just clickable screenshots, right

You can have a working prototype ready in minutes with Perplexity Computer. no coding required no additional tools no switching between apps here’s what I created 👇

2M ARR claim in 3 weeks. $1 first month pricing. Do the math: every $1 customer is counted as $300 ARR by annualizing the full price. This is legal. It is also a lie and quite dishonest. VCs know. They fund it anyway because the metrics close the next round.

@SpiralEye_ wait what actually changed day to day between gpt and claude for you?

My life using GPT vs. My life using Claude

@Shahzaib0x @idolly_AI @solana this is just chatbots selling crypto. how does idolly_ai avoid the spam agent problem?

Watched this and honestly… this feels like the next evolution of creator economy. With @idolly_AI building on @solana, it is not just about AI agents it is about turning your online identity into something that actually works for you 24/7. From what I have seen, the idea of an agent that can post, create visuals and engage like you is powerful. But what really stood out to me is the ownership layer launching your own token and building a community where value flows back to both creator and fans. Also, the integration with Metaplex for on-chain identity adds a level of trust that most AI tools lack right now. Feels like we’re moving from “content creators” to “digital economies run by personalities.” Curious to see how far this can go.

@NuriGems wait how does agent discovery work without a centralized registry or trust layer?

Anvita Flow has been unveiled at Cannes, France! 🚀 During the Real Up Cannes Summit, introduced the new infrastructure for AI agent collaboration. Personal agents will be able to register, discover capabilities, collaborate autonomously (A2A), and exchange value instantly thanks to the x402 micropayment protocol. -Agent Store with ready-made modules -One-click orchestration -Compatible with frameworks like OpenClaw and Claude Code -Full security for data, identity, and transactions Together, the Anvita brand is born: • Anvita Flow → for individual users (agencyization of on-demand services) • Anvita TaaS → Tokenization-as-a-Service for institutions The future of the Agent-to-Agent economy is taking shape! techbullion.com/anvita-flow-la…

if notion ai disappeared tomorrow, most teams wouldn’t notice after a week. that’s the tell. so why are people paying for it instead of better workflows or actual automation?

compared to just using chatgpt with pasted context, notion ai is slower and less flexible. prompts are constrained and you can’t iterate deeply without starting over.

the core issue: it doesn’t understand your workspace. it rewrites text fine, but it can’t reason across pages, tasks, or databases in a useful way. it’s a fancy text box glued onto notion.

@goodalexander one question, how exactly does claude code underwrite credit risk without eating lawsuits?

so private credit is: A. part software that is automatable via Claude Code B. part low end consumer that will be destroyed by high oil prices and high rates C. a bundle of mystery meat facing massive redemptions that cannot be met D. $3.5 trillion of notional seems non ideal

@TheGeorgePu wait what makes claude code the most popular? by users or just twitter vibes?

Claude Code is the most popular AI coding tool in the world. One packaging error and the entire playbook became public. If a $380 billion company can't protect its tooling. What makes you think your AI wrapper is defensible?