Vikas Pandey

Vikas Pandey@buidloor · 2026/05/03赛道影响

@alex_scharrer Wow. Cant wait to hear more @alex_scharrer. Been planning myself to travel this year. We got to buckle up, fosho.

101208

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Vikas Pandey@buidloor · 2026/05/02赛道影响

@alex_scharrer @nasib_fathi @near_intents @near_ai @NEARProtocol Pretty impressive to learn the institutional perspective @alex_scharrer 🦾. Great stuff.

00268

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Vikas Pandey@buidloor · 2026/05/01赛道影响

@nasib_fathi @alexscharrer @NEARFoundation great insights @alex_scharrer . This is a must-watch for those looking to understand how institutions are approaching AI adoption.

20249

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Vikas Pandey@buidloor · 2026/03/19赛道影响

Most agent frameworks are being layered onto infrastructure built for humans such as browsers, wallets, auth flows designed around human latency and human judgment. However. Agents don’t work that way. They need: - Deterministic payment rails (not “sign txns and wait”) - Machine-readable auth, not OAuth flows with consent screens - Coordination that don’t assume a human in the loop Building that at the L1 level is the correct architectural direction. If you’re a builder thinking seriously about the agentic economy, this hackathon is worth your time. Three tracks, real infrastructure, real constraints to push against. The agents that matter won’t just think. They’ll transact. Join us and Build with @GoKiteAI

005172

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Vikas Pandey@buidloor · 2026/03/17政策影响

A pattern I keep seeing in agentic systems: You automate the work. The work comes back in a new shape. My current Claude stack: - Test agent - needs debugging - Security agent - needs upgrades & patching - Social agent requires constant tweaking The bottleneck didn't disappear with agentic flow, it moved upstream. This is what I have been arguing in my Agentic Engineering Patterns series : The shift isn't from more work to less work. It's from execution to architectural judgment. Directing systems is the new doing.

3012153

AI：偏政策推进，强调合规落地与规则明晰，对监管主线更敏感。

Vikas Pandey@buidloor · 2026/03/17观点输出

@itsalexey Really cool @itsalexey

00113

AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。

Vikas Pandey@buidloor · 2026/03/16赛道影响

Congratulations to all ETH Denver winners. Some really nice projects were built on @GoKiteAI. Check them out below and also at our youtube channel. More to some soon during our Global Hackathon, end of this month. Stay tuned…

105177

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Vikas Pandey@buidloor · 2026/03/14赛道影响

Been digging into world models lately, the scale of what’s happening here is kinda wild : - @AMIlabs just raised $1.03B seed at $3.5B valuation to build world models, one of the largest seed rounds ever in AI - Some world model datasets exceed 500M+ video frames to learn physical dynamics like motion, collisions & object interaction - RoboNet dataset alone contains 15M robot interaction frames across 7 robot platforms for world modelling. - NVIDIA’s physical-AI work is reportedly using 20M hrs of video to teach models how the real world behaves

00274

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。

Vikas Pandey@buidloor · 2026/03/13政策影响

Just dropped Part #3 of my “Agentic Engineering Patterns” series : - "When Prompts Become Interfaces." In this post, I dive into: - How prompts are transforming from simple inputs into dynamic, intent-driven interfaces in the agentic AI world. - From command-lines and GUIs to natural language prompts that let users (and agents) declare needs like "Generate a quarterly sales report”, less menus or clicks required. Some Key insights: - Prompts add a third layer alongside GUIs (for humans) and APIs (for machines), demanding "prompt-friendly" designs with rich metadata for discoverability. - Challenges include intent interpretation, ambiguity handling (e.g., clarifying vague requests), mapping to actions via APIs, and building guardrails for safety. Good practices: - Decouple intent from action : parse prompts, validate, then execute securely to keep systems transparent and reliable. This shifts software from rigid tools to intuitive assistants, essential for scalable architectures. If you're building AI systems, this pattern is a game-changer. Read here: https://t.co/xKYEr4uUFJ

004126

AI：偏政策推进，强调合规落地与规则明晰，对监管主线更敏感。

Vikas Pandey@buidloor · 2026/03/12赛道影响

Was reading more about this today (Huge Yann LeCun fan) - AMI reportedly raised $1B+ at a $3.5B valuation, one of the largest seed rounds in AI history. Their thesis is interesting - move beyond LLMs and build world models : Targeting robotics, manufacturing, aerospace, and autonomous systems. One key insight stood out to me: - Large Language Models understand text, not the world. - They learn patterns in language, but concepts like 3D space, physics, causality, or persistent environments don’t naturally exist in their training data (which uses transformer architecture). - Everything gets flattened into tokens. That’s why world models will emerge further, where systems are designed to learn how the world behaves, and not how we humans describe it. TLDR ; Language models predict the next word. World models try to predict the next state of reality.

007115

AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。