Legendary AI engineer and educator Andrej Karpathy recently blogged about his experiences building, deploying, and monetizing a vibe-coded web app called MenuGen. Let's dig into the challenges he faced and learn what we as AI designers can do to make life better for the Andrejs of the world. About Zeke Sikelianos Zeke's been building developer tools at companies like Heroku, npm, GitHub, and Replicate for over ten years. He cares deeply about simple and tasteful developer experiences, and thinks the world o ...

[Music] All right. Hey everyone. Uh, my name is Sharief. I'll be talking to you about demos and why I think demos are probably the most important thing in the world right now. Um, I'm the founder of Lexica. We're working on generative models, specifically image models. Um, but I kind of want to just talk to you about something a bit more than just models themselves. Um, even more than demos. I kind of just want to talk to you about curiosity. Um, there was a famous French mathematician Poare. He said at the ...

AI Development Capabilities - The industry is focusing on bringing AI development capabilities through Copilot, starting with code completion and moving towards chat interactions for complex prompts and multi-file changes [1] - Agent mode enables complete task execution with deep interaction, allowing for building apps or refactoring large codebases [2] - Agent mode can interpret readme files, including project structure, environment variable configurations, database schemas, API endpoints, and workflow graphs (even as images), to implement tasks [3][4][5] Model Context Protocol (MCP) - MCP is an open protocol (API for AI) that allows LLMs to connect to external data sources for general or account-specific information [9] - VS Code can be configured to use specific MCPs, allowing Copilot to select the appropriate MCP for a task and connect to it, whether local or remote [11][12] - Developers need to grant permission for Copilot to connect to MCPs, ensuring data access is controlled [20] - GitHub has its own MCP server, enabling actions like committing changes to a new branch and creating pull requests directly from the IDE [26][31] Workflow and Best Practices - Copilot Instructions, a specially named file, can be used to pre-inject standards and practices into every prompt, such as code style guidelines and security checks [28][29][30] - Including a change log of everything the agent has done provides a clear record of each step taken [30]

And uh but hopefully at the end I'll uh relieve that stress a little bit with some ideas and solutions for you. So I need everybody to just think for a second, reflect on the past 45 days and think about all the possible things that have gone on in our industry and all the product launches. Let me highlight just a few for you.But none of that matters. None of it matters unless you answer this question. What do I build and why will it win.Your competitive advantage will come from what is uniquely yours. thes ...

Agentic Economy & MCP Standard - The agentic economy is emerging, where AI agents can interact, find counterparts, and purchase services from other agents, businesses, or tools [4] - MCP (Message Communication Protocol) is becoming a standard for agentic interaction, dominating the space compared to Open API and Google's A2A [8][9] - Tool discovery, a key feature of MCP, allows agents to dynamically discover and use tools based on the workflow, differentiating it from Open API [7][8] - A centralized marketplace of MCP services, like APIFY, can provide access to various services with a single API token, enabling rapid scaling of the ecosystem [12] APIFY's Role & Marketplace - APIFY is a marketplace of 5,000 tools (actors), primarily data extraction tools, with a community of creators who monetize their tools [4] - Actors are self-contained software units with defined input and output, facilitating easy integration with other systems [4][5] - APIFY has integrations with workflow automation tools and MCP, enabling AI agents to call actors from the marketplace [6][7] - APIFY enables publishing and monetization of tools or agents, providing access to a broad ecosystem of developers and visibility [23][24] Challenges & Future - Agents currently rely on human developers for access to tools and services, hindering their ability to autonomously find and purchase services [10][11] - Trust between agents and tools is a key open question, as is the overall value and reliability of autonomous tool discovery [25][26][27] - The company paid out over $4 million to creators last month, with actors generating over $500,000 per month, indicating rapid ecosystem growth [23]

MCP Ecosystem & Specification - The Model Context Protocol (MCP) ecosystem is still in its early stages, with significant room for growth and development [2][3] - The industry emphasizes the importance of adopting the full MCP specification to unlock rich, stateful interactions between agents [9] - The industry acknowledges a gap in MCP implementation, with a tendency to treat it as just another API wrapper [5] - Technical barriers, including missing support in clients, SDKs, documentation, and references, contribute to the limited adoption of the full MCP spec [6] - The industry highlights the need for developers to stay updated with the latest MCP specification and provide feedback on draft features [29] Tools & Dynamic Discovery - Tools are the most immediately successful aspect of MCP, but overuse can lead to quality problems and AI confusion [7][11][12] - Dynamic tool discovery allows servers to provide context-aware tools, enhancing the user experience [16][17][18] - VS Code offers user controls like per-chat tool selection and user-defined tool sets to manage tool complexity [13][15] Resources & Sampling - Resources provide a semantic layer for exposing files and data to both the LLM and the user, enabling more dynamic and stateful interactions [19][20] - Sampling allows servers to request LLM completions from the client, enabling progressive enhancement and interesting functionalities [22][23][24] Developer Experience & Community - The industry recognizes the need for improved developer experience when working on MCP servers, including debugging and logging [26] - VS Code offers a dev mode with debugging capabilities for MCP servers, simplifying the development process [26][27][28] - A community registry is being developed to facilitate the discovery of MCP servers [32]

What does it take to go from blank page to live enterprise voice agent in 100 days? That’s the challenge we took on with Fin Voice at Intercom. Enterprise customer service demands high-quality, reliable voice interactions - but delivering that fast means wrestling with tough problems like latency, hallucinations, voice quality, and answer accuracy. We rapidly evaluated and integrated a full voice stack - including transcription, language model, text-to-speech, retrieval-augmented generation, and telephony - ...

All right, I need everybody to take a deep breath here because um I'm about to stress you out. And right, this is just one little microcosm of the entire tech industry, but if you look around at all the different categories of software right now, the same exact thing is happening. And I haven't even mentioned the horde of startups, wellunded startups, uh that are getting funded in every single one of these spaces as well.And among all of this chaos, we have companies that are essentially collapsing in month ...

Generative Media Platform & Market Overview - File.ai 将自身定义为一个生成式媒体平台，专注于视频、音频和图像的生成 [1] - 生成式媒体正在改变社交媒体、广告、营销、时尚、电影、游戏和电子商务等行业，最终将影响所有内容 [10] - 广告行业预计将成为首批大规模受到生成式媒体影响的行业之一，行业规模预计将会扩大 [13] AI Model Development & Trends - 边缘计算的创作边际成本正在接近于零，但故事叙述和创造力仍然至关重要 [8][9] - 视频模型的使用率正在快速增长，从10月初的几乎为零增长到2月份的18%，并且持续增长，目前约为30% [25][26] - 视频模型预计将比图像生成市场大 100 到 250 倍，因为视频模型计算密集程度是图像的 20 倍，互动性是图像的 5 倍，并且将影响更多行业 [27] - 视频生成技术将朝着更快、更便宜的方向发展，最终实现实时视频生成，这将对用户互动方式产生重大影响，模糊游戏和电影之间的界限 [31] - 图像模型也在不断改进，例如 Flux context 和 GPT4o 引入了新的编辑功能和更好的文本渲染功能，为行业开辟了更多用例 [34] Applications of Generative Media - 个性化广告是生成式媒体的一个重要应用方向，可以针对不同的人口统计群体快速生成大量不同版本的广告，或者根据用户的浏览行为动态生成广告 [15] - 电子商务是生成式媒体的另一个重要应用领域，特别是虚拟试穿技术，许多零售商和初创公司都在采用这项技术 [21][22] - AI 正在帮助创建互动和个性化的体验，例如 A24 电影《内战》的互动广告活动，用户可以将自己的自拍照放在时代广场的玩具士兵上 [18][19]

Search & Discovery Challenges in Grocery E-commerce - Instacart faces challenges with overly broad queries (e.g., "snacks") and very specific, infrequent queries (e.g., "unsweetened plant-based yogurt") due to limited engagement data [6][7] - Instacart aims to improve new item discovery, similar to the experience of browsing a grocery store aisle, but struggles due to lack of engagement data [8][9][10] - Existing models improve recall, but maintaining precision, especially in the long tail of queries, remains a challenge [8] LLM-Powered Query Understanding - Instacart utilizes LLMs to enhance query understanding, specifically focusing on query to category classification and query rewrites [10][11][12] - For query to category classification, LLMs, when augmented with top converting categories as context, significantly improved precision by 18 percentage points and recall by 70 percentage points for tail queries [13][21] - For query rewrites, LLMs generate precise rewrites (substitute, broader, synonymous), leading to a large drop in queries with no results [23][24][25][26] - Instacart pre-computes outputs for head and torso queries and caches them to minimize latency, while using existing or distilled models for the long tail [27][28] LLM-Driven Discovery-Oriented Content - Instacart uses LLMs to generate complementary and substitute items in search results, enhancing product discovery and user engagement [31][34] - Augmenting LLM prompts with Instacart's domain knowledge (e.g., top converting categories, query annotations, subsequent user queries) significantly improves the relevance and effectiveness of generated content [39][40][41] - Instacart serves discovery-oriented content by pre-computing and storing content metadata and product recommendations, enabling fast retrieval [42][43] Key Takeaways & Future Directions - Combining LLMs with Instacart's domain knowledge is crucial for achieving topline wins [47] - Evaluating content and query predictions is more important and difficult than initially anticipated [47][48] - Consolidating multiple query understanding models into a single LLM or SLM can improve consistency and simplify system management [28]