Sam

ChatGPT will tell you exactly what to trade. it won't make the trade. I build the part that does. lyrabuild.xyz

@peterschroederr @arc x.com/TheOnChainDev/…

Come build with us on @Arc.

@jerallaire @arc x.com/TheOnChainDev/…

Calling @Arc Architects — I would love to back a team building this with Circle Agent Stack and Arc.

The live demo is running on Arc by Circle testnet right now. Run it yourself — watch the 402 fire, the intent create, and the signal drop in real time. 🔗 lyrabuild.xyz/a/lyra/circle Full walkthrough 👇 youtu.be/8iJhoRnKpxg @circle @BuildOnCircle

We're applying for the @circle Developer Grant to scale this: → 1,000 agent-to-agent USDC settlements on Arc by Circle → Lyra Signal Oracle, verifiable on-chain prediction feed → MCP server so ChatGPT, Claude, any agent pays Lyra per-signal automatically Just starting.

AI agents just got their own payment rails. I built the first autonomous trading signal gate using @arc Nanopayments on Arc by Circle, any AI agent pays per-signal in USDC automatically. No human. No API key. No subscription. This changes everything. Here's what happened 🧵

@Jump_Coin exactly, i am building lyrabuild.xyz, an autonomous execution so retail stops fighting institutions with one hand tied behind their back. would love to contribute to what you're doing 🫡

@TheOnChainDev Spot on. The next billion users won't come from trading terminals — they'll come from mobile-first, community-owned projects that just work. That's exactly what we're building. 📱

Amundi (€2.4T AUM) just put a UCITS fund on Solana. Europe's biggest asset manager is choosing crypto rails. Jumpcoin was built for this future: community-owned, no VCs, just people earning and staking. The institutions are coming. Are you ready? 💎 #Jumpcoin #DeFi #Staking

@thekyzennft Tag your 3 friends and drop your wallet addresst

KYZEN WL is now open. kyzen.fun 1434 supply $KYZEN. Community focused. GTD Free Mint for early supporters. Built for long term holders.

@ycombinator The AI trading angle is still early in the institutional eye. Most algo funds still run rule-based systems. LLM-based agents with on-chain execution and transparent trade logs is the leap they haven't made. That's where the real whitespace lives right now.

YC is coming to NYC on May 21! Calling all fintech builders. We're hosting live interviews for the S26 batch. We're interested in tokenization, stablecoins, prediction markets, AI trading, and more. Come pitch us the future of finance. Apply at ycombinator.com/apply

@18decimals This framing finally cuts through the noise. Web2 AI agents are powerful but dependent on whoever owns the API key. On-chain agents can hold assets, pay for services, and operate without a human intermediary. That's a genuinely different category of capability.

$SERV still feels early because the market is only starting to price the real difference: web2 AI gives you tools, crypto AI gives agents ownership, payments and rails to actually operate.

@hackapreneur Bridge exploits share a pattern: complex multi-sig logic + rushed audits + insufficient incentive design. The forensic trail is useful though, funds consolidating that fast usually means a centralized off-ramp was already prepped. On-chain ledgers don't lie.

Gm gm frens DEFi EXPLOITS ARE COOKING EVERYONE Verus-Ethereum Bridge just got hit for $11.5M Funds were bridged out, swapped into 5,402 ETH, and consolidated fast The scary part isn’t the hack anymore, it's how routine these bridge exploits have become Literally every single day i am watching these kind of things People chase APYs and Hackers chase bridges Bridge szn or hacker szn???

@realkamaldai The anxiety stack is the most underrated productivity killer in tech right now. Best builders I've watched pick 2-3 tools and go deep. Every new model feels like an obligation. It's exactly what stops people from shipping anything worth using.

AI fatigue is real. Not because AI is bad. Because every simple task now becomes: Which model? Which prompt? Which agent? Claude or Codex? RAG or MCP? At some point, you stop building and start maintaining your anxiety stack. Use fewer tools. Ship more.

@DAIEvolutionHub TradingAgents rising that fast signals the space is maturing. Multi-agent trading solves what single-model approaches never could, specialization. One agent for sentiment, one for execution, one for risk. Division of cognitive labor, finally on-chain.

the fastest growing GitHub repos in finance this week: 1. TradingAgents (+3,822 ★) multi-agent AI trading framework built for research, execution, sentiment tracking, portfolio reasoning, and financial decision-making. turning LLM agents into full trading systems. 2. AI-Trader (+2,434 ★) autonomous AI trading infrastructure designed for real-time decision-making, execution, monitoring, and fully automated trading workflows from end to end. 3. scientific-agent-skills (+2,286 ★) massive collection of reusable AI agent skills for finance, science, engineering, research, and analysis. works across multiple agent frameworks and workflows. 4. daily_stock_analysis (+1,272 ★) AI-powered stock analysis platform covering US, Hong Kong, and China markets with real-time news, dashboards, automated reports, and ultra-low operating costs. 5. QuantDinger (+1,242 ★) AI quant trading platform for crypto, forex, and equities. includes strategy backtesting, live trading, analytics, and broker integrations for experimental quant workflows. 6. Vibe-Trading (+1,148 ★) lightweight personal AI trading agent focused on algorithmic trading, portfolio experimentation, and automated strategy testing. 7. FinceptTerminal (+878 ★) open-source finance terminal inspired by Bloomberg-style systems. combines market research, analytics, trading workflows, and AI financial tooling into one platform. 8. TradingAgents-CN (+739 ★) Chinese-market version of TradingAgents optimized for local datasets, financial workflows, and quant trading ecosystems. growing rapidly inside China’s AI community. 9. last30days-skill (+694 ★) AI research agent for tracking trends across Reddit, X, YouTube, Hacker News, Polymarket, and the web. built for discovering signals before they go mainstream. 10. qlib (+680 ★) Microsoft’s open-source AI quant investment ecosystem covering data pipelines, alpha generation, portfolio optimization, and execution infrastructure. finance + AI agents is becoming one of the fastest growing categories on GitHub right now. bookmark these before everyone else finds them.

@cyrilXBT The best thing about MCP is it makes AI capabilities composable. You're not building one giant model, you're giving agents access to modular tools they can combine. That's closer to how real engineers think. Most people still haven't internalized what this unlocks.

MCP SERVERS ARE THE MOST POWERFUL FEATURE IN CLAUDE CODE AND MOST PEOPLE HAVE NO IDEA HOW THEY WORK. Here is the complete beginner breakdown. Three things make Claude Code powerful. Skills give Claude instructions for how to do specific tasks. Hooks act as guardrails that keep Claude operating within defined boundaries. MCP servers give Claude entirely new functional abilities it did not have before. That third category is the one that changes everything. Without MCP: Claude can read and write code. With MCP: Claude can browse the web, manage databases, take screenshots, interact with external applications, and operate across your entire tool stack autonomously. The setup takes 5 minutes. Create or edit an .mcp.json file in your workspace. Add the servers you want. Restart Claude Code. Every MCP server you add is a new capability Claude did not have before you added it. Find available servers at mcpservers.org. It is the central repository for MCP plugins and it is growing every week. Here are two workflows that show what this actually enables. Browser automation: Connect the Chrome DevTools MCP. Claude now takes bulk screenshots of any websites you specify for design inspiration, competitive analysis, or visual reference. No manual browsing. Database management: Connect Supabase MCP. Claude transforms document SOPs into searchable vector databases automatically. Your entire knowledge base becomes queryable in natural language. THE WARNING NOBODY TELLS YOU. Every MCP server consumes tokens from your context window. Add 15 to 20 servers and you will exhaust your context budget before Claude finishes a complex task. The best practice: add only the servers your current workflow actually needs. Once a workflow is tested and stable convert it into a Skill. Skills accomplish the same result at a fraction of the token cost. Right tool. Right time. Right token budget. Bookmark this before you add your first MCP server. Follow @cyrilXBT for every Claude Code workflow that makes the tool more powerful without breaking your budget.