Agent Architecture & Functionality - The Langchain deep research agent is highly configurable and open source, allowing for customization to specific use cases [1] - The agent operates in three main phases: scoping the problem, research, and report writing [3] - The research phase utilizes a supervisor to delegate tasks to sub-agents for in-depth research on specific subtopics [4] - Sub-agents use a tool calling loop, which can be configured with default or custom tools (like MCP servers) for searching flights, hotels, etc [17][18] - A compression step is used by sub-agents to synthesize research findings into comprehensive mini-reports before returning to the supervisor, mitigating context window overload [21][23] - The supervisor analyzes findings from sub-agents to either complete research or continue with follow-up questions [25] - Final report generation is done in a single shot using all condensed research findings [5][27] Implementation & Configuration - The agent is built on Langraph and can be run locally by cloning the Open Deep Research repository [29] - Configuration involves setting API keys for models (default OpenAI) and search tools (default Tavily) [30] - Langraph Studio can be used for iterating and testing the agent with different configurations [32] - The agent is highly configurable, allowing users to choose between default or model provider native search tools, connect to MCP servers, and select models for different steps [33][34] Application & Output - The agent can be used for complex research tasks, such as planning a trip, by iteratively calling tools and searching the web [2] - The agent provides a final report with an overview, flight options, transportation options, accommodation options with booking links, a sample itinerary, and a list of sources [36] - Open Agent Platform provides a UI to configure and try out the research agent without cloning the code [37]

Partnership & Integration - Coinbase partnered with Perplexity AI, indicating a strategic move beyond data provision [1] - The partnership aims to provide traders with an AI edge, suggesting enhanced trading capabilities [1] - The collaboration lays the groundwork for full Large Language Model (LLM) integration into the crypto economy [1] Potential Impact - The integration could significantly alter the crypto landscape, implying transformative changes [1]

If you found it insightful, reshare it with your network.Find me → @_avichawlaEvery day, I share tutorials and insights on DS, ML, LLMs, and RAGs.Avi Chawla (@_avichawla):Now you can supercharge your terminal with MCP servers (open-source).MCP CLI lets you interact with local and remote MCP servers, built with a rich UI, and full LLM provider integration.You can run tools, manage conversations, or automate workflows directly from your https://t.co/FME7lArlTZ ...

Core Principles of Agent Building - The industry emphasizes rethinking agent development from first principles, applying established software engineering practices to build reliable agents [11] - The industry highlights the importance of owning the control flow in agent design, allowing for flexibility in managing execution and business states [24][25] - The industry suggests that agents should be stateless, with state management handled externally to provide greater flexibility and control [47][49] Key Factors for Reliable Agents - The industry recognizes the ability of LLMs to convert natural language into JSON as a fundamental capability for building effective agents [13] - The industry suggests that direct tool use by agents can be harmful, advocating for a more structured approach using JSON and deterministic code [14][16] - The industry emphasizes the need to own and optimize prompts and context windows to ensure the quality and reliability of agent outputs [30][33] Practical Applications and Considerations - The industry promotes the use of small, focused "micro agents" within deterministic workflows to improve manageability and reliability [40] - The industry encourages integrating agents with various communication channels (email, Slack, Discord, SMS) to meet users where they are [39] - The industry advises focusing on the "hard AI parts" of agent development, such as prompt engineering and flow optimization, rather than relying on frameworks to abstract away complexity [52]

AI Development History - The AI debate in the 2010s was framed by two books: one predicting superhuman intelligence and the other focusing on AI as surveillance tech [1][2] - Reality, exemplified by LLMs and ChatGPT, fell between these extremes, aligning with the traditional definition of AI as passing the Turing test [2][3] - ChatGPT's ability to pass the Turing test is considered a significant achievement [3]

Company Overview - TME is committed to the healthy development of China's online music industry[7] - TME has a large user base with 576 million online music MAUs[13] and 90 million social entertainment mobile MAUs[15] in 3Q2024 - TME boasts an extensive content library with over 200 million music and audio tracks[14, 20] and 480K+ indie musicians[15] - TME's total cash, cash equivalents, term deposits, and short-term investments reached RMB 3604 billion[15, 18, 49] Business Overview - TME has partnerships with hundreds of domestic and international music labels[20] - TME is expanding LLM capabilities, AIGC tools & full-suite of resources and services to streamline content production[21] - TME cultivates and empowers indie musicians & original music through Tencent Musician Platform[22] Financial Highlights - TME's online music monthly ARPPU was RMB 108 in 3Q24, a 49% year-over-year increase[37] - TME's revenue from music subscriptions reached RMB 384 billion in 3Q24, a 203% year-over-year increase[37] - TME's gross margin was 426% in 3Q24, a 69 percentage point year-over-year increase[37] - TME's Non-IFRS net profit was RMB 194 billion in 3Q24, a 291% year-over-year increase[37]

AI Coding Agents: Efficacy and Usage - Coding agents are substantively useful, though opinions vary on their best practices and applications [1] - The number one mistake people make with coding agents is using them the same way they used AI coding tools six months ago [1] - The evolution of frontier model capabilities drives distinct eras in generative AI, influencing application architecture [1] Design Decisions for Agentic LLMs - Agents should make edits to files without constant human approval [2] - The necessity of a thick client (e.g., forked VS Code) for manipulating LLMs is questionable [2] - The industry is moving beyond the "choose your own model" phase due to deeper coupling in agentic chains [2] - Fixed pricing models for agents introduce perverse incentives to use dumber models [2] - The Unix philosophy of composable tools will be more powerful than vertical integration [2] Best Practices and User Patterns - Power users write very long prompts to program LLMs effectively [4] - Directing agents to relevant context and feedback mechanisms is crucial [5] - Constructing front-end feedback loops (e.g., using Playwright and Storybook) accelerates development [6] - Agents can be used to better understand code, serving as an onboarding tool and enhancing code reviews [9][11] - Sub-agents are useful for longer, more complex tasks by preserving the context window [12][13]

If you found it insightful, reshare it with your network.Find me → @_avichawlaEvery day, I share tutorials and insights on DS, ML, LLMs, and RAGs.Avi Chawla (@_avichawla):A Python decorator is all you need to trace LLM apps (open-source).Most LLM evals treat the app like an end-to-end black box.But LLM apps need component-level evals and tracing since the issue can be anywhere inside the box, like the retriever, tool call, or the LLM itself. https://t.co/dWXyJb3DNs ...

AI Agent Evaluation & Observability - The industry emphasizes the necessity of observability in AI development, particularly for evaluation-driven development [1] - AI trustworthiness is a significant concern, highlighting the need for robust evaluation methods [1] - Detecting problems in AI is challenging due to its non-deterministic nature, making traditional unit testing difficult [1] AI-Driven Evaluation - The industry suggests using AI to evaluate AI, leveraging its ability to understand and identify issues in AI systems [1] - LLMs can be used to score the performance of other LLMs, with the recommendation to use a better (potentially more expensive or custom-trained) LLM for evaluation than the one used in the primary application [2] - Galileo offers a custom-trained small language model (SLM) designed for effective AI evaluations [2] Implementation & Metrics - Evaluations should be integrated from the beginning of the AI application development process, including prompt engineering and model selection [2] - Granularity in evaluation is crucial, requiring analysis at each step of the AI workflow to identify failure points [2] - Key metrics for evaluation include action completion (did it complete the task) and action advancement (did it move towards the goal) [2] Continuous Improvement & Human Feedback - AI can provide insights and suggestions for improving AI agent performance based on evaluation data [3] - Human feedback is essential to validate and refine AI-generated metrics, ensuring accuracy and continuous learning [4] - Real-time prevention and alerting are necessary to address rogue AI agents and prevent issues in production [8]

Voyage AI & MongoDB Partnership - Voyage AI was acquired by MongoDB approximately 3-4 months ago [1] - The partnership aims to create a single data platform for embedding, re-ranking, query augmentation, and query decomposition [29][30][31] AI-Powered Search & Retrieval - AI-powered search finds related concepts beyond identical wording and understands user intent [7][8][9] - Embedding quality is a core component, with 95-99% of systems using embeddings [12] - Real-world applications include chatting with codebases, where evaluation is crucial to determine the best embedding model and LLM for the specific application [14][15] - Structured data, beyond embeddings, is often necessary for building powerful search and retrieval systems, such as filtering by state or document type in legal documents [16][17][18] - Agentic retrieval involves feedback loops where the AI search system is no longer just input-output, but can expand or decompose queries [19][20] Future Trends - The future of AI-powered search is multimodal, involving understanding images, text, and audio together [23][24][25] - Instruction tuning will allow steering vectors based on instructions, enabling more specific document retrieval [27][28]