PodcastMLOps at the CrossroadsPushed backLLMOps should be treated as a distinct specialization with specialized tools rather than only as an extension of existing MLOps.19:55
Threads / Agents in production / 2024
Agents in production in 2024
39 sessions
PodcastMLOps at the Crossroads
PodcastMicro Graph Transformer Powering Small Language ModelsClaimJon Cooke says business teams need to define the desired outcome clearly before an agent or analytics system can determine how to achieve it.40:57
A Survey of Production RAG Pain Points and SolutionsClaimRAG response-quality problems can result from missing knowledge-base context, failed retrieval, failed reranking, extraction failures, incorrect output formats, incorrect specificity, or incomplete answers.14:04
PodcastBecoming an AI EvangelistPushed backDemetrios Brinkmann says agents are not yet ready, while Alex Volkov argues that they are improving and should not be expected to remain at their current level forever.1:21
From Robotics to AI NPCsClaimAn AI NPC is an embodied AI agent that can respond to a person, move its body, and perform requested tasks.0:56
Let's Build a Website in 10 Minutes with GitHub CopilotClaimGitHub Copilot is an artificial-intelligence code-completion tool that helps developers write code faster.1:19
PodcastManaging Small Knowledge Graphs for Multi-agent SystemsPushed backTom Smoker disputes the idea that one natural-language instruction should be trusted to make a multi-agent system complete a task correctly every time.36:41
PodcastNavigating the AI Frontier: The Power of Synthetic Data and Agent Evaluations in LLM DevelopmentPushed backBoris Selitser argued that online intervention is often undesirable for complex agents because they may recover or find useful workarounds that developers did not anticipate.13:42
PodcastML and AI as Distinct Control Systems in Heavy Industrial SettingsClaimRichard Howes compares a control philosophy to context supplied to a language model: it gives data scientists the business process, diagrams, operating principles, constraints, and optimization objectives before they analyze data.42:15
PodcastEvaluating the Effectiveness of Large Language ModelsClaimAniket Singh is moving toward research on production applications, multimodal systems, and multi-agent workflows because reliability is a major problem when using language models in production.21:10
Reading groupIntegrating Knowledge Graphs & Vector RAG for Efficient Information ExtractionPushed backValdimar Eggertsson argued that recall should generally be prioritized over precision, while Sonam Gupta said excessive context can reduce model performance and make precision important.14:48
PodcastMaking Your Company LLM-nativeClaimPampa Labs experiments with office agents for tasks such as collecting food orders and splitting team expenses, then checks whether the agents save more time than they require.7:40
Building Hyper-Personalized LLM Applications with Rich Contextual DataClaimPersonalization requires giving an AI model enough context to know the user, rather than relying only on the model's general expertise.5:02
PodcastHow Agentic Workflows Will Change EverythingPushed backGiving an agent raw web access is disputed as an unsafe default because it provides too much agency without sufficient constraints.16:45
The Coming Revolution of AI AgentsClaimJazmia Henry is building open-source models for agentic AI with better context.1:03
LLMs to agents: The Beauty & Perils of Investing in GenAIPushed backThe panel discusses whether AI agent startups should be classified as SaaS companies or infrastructure companies and concludes that the answer depends more on the customer, budget, buying behavior, and distribution than on a fixed category.29:04
Why We Need More Data Science PodcastsClaimDelphina is betting that AI-powered agents can make data science faster by helping with exploratory data analysis, data munging, modeling and the iterative process of experimentation.16:42
Cleric AI SRE: Towards Self-healing Autonomous SoftwareClaimCleric combines a reasoning engine, production tools, a knowledge graph, and memories of previous interactions and solutions.6:47
AI Agents: The Future of Productivity, or Just a Fad?Pushed backSam Partee says retrieval systems are not agents and that an agent must be able to call tools.4:50
Few Shot Code Generation to Autonomous Software Engineering AgentsPushed backSome academic feedback characterized SWE-bench as too difficult, especially for an execution-free, non-agent setting.16:07
How AI Agents Are Transforming Data AnalyticsPushed backThe host raises a tension between agent performance and enterprise permissioning, while Ines Chami says permissioning should be inherited from the existing source systems rather than rebuilt.18:21
How AI Will Change Gaming ForeverClaimPietro Gagliano defines AI agents in entertainment as autonomous systems used in production, including systems that are not based on large language models.4:51
How AI Agents Will Change Customer SupportPushed backNeil Laia says the industry does not yet have uniform conceptual frameworks for building AI-agent systems comparable to those used for large distributed systems.15:59
We're Using AI Agents at Work (and it's amazing)ClaimProsus built an internal assistant called Toan to help colleagues experiment with generative AI and use agents in their work.2:34
AI Agents Are Revolutionizing E-CommerceClaimOLX Magic uses agentic flows together with OLX’s catalog search.4:53
The Open Source AI Coding RevolutionClaimOpenHands aims to solve complete software issues autonomously rather than only providing synchronous code completion.1:53
How to Make AI Agents that ACTUALLY WORKPushed backPatrick Marlo disputes the view that a production agent is mainly the underlying language model.4:44
Hundreds of Users Love Our Data Analyst AI AgentClaimAgents can understand questions, find relevant data sources, generate and execute SQL, recover from failures, and provide visualizations, but they do not work reliably out of the box.3:01
Knowledge as a ServiceClaimStack Overflow's main AI-related problems are the loss of new human-created training content, AI systems reaching a complexity limit, and users' difficulty trusting AI tools.4:04
Hugging Face Cofounder on AI Agents, LLMs and Open SourceClaimThe term AI agent covers very different use cases, including coding tools, business-process automation and robots.7:56
Building Reliable AgentsClaimEno Reyes defines an agentic system by three characteristics: planning, decision-making, and environmental grounding.2:57
Building Replit Agent - Hard Lessons LearnedPushed backJames rejected the idea that public evaluations were sufficient for judging an agent built for a specific product and user group.15:25
Maximize Your Productivity with LLMs: Task Utility ExplainedClaimJulia Kiseleva says AgentEval is intended to help developers understand the utility an agent application may provide to users without requiring ground-truth data.5:11
How to Create a Multi-Agent AI System in JavaScriptPushed backFrontend engineering jobs will not disappear, but the number of people needed may decrease and roles may shift toward directing agents.14:45
Simulation Techniques for AI Agents from Self-DrivingPushed backBrooke Hopkins disagreed with the view that a human should always remain in the loop, arguing that teams should build toward full autonomy.17:43
Why Agents Are Stupid & What We Can Do About ItPushed backDan Jeffries disputes the idea that a mixture of agents or models magically solves cascading errors.26:36
Autonomous Multi Agent AI SystemsClaimNatan defines an agent as an autonomous unit that can perform tasks and make decisions, using tools and a system prompt describing what it should do.1:51
Why Pydantic AI is the Future of AI AgentsPushed backSamuel Colvin disputed the idea that Pydantic AI should replace other agent frameworks for everyone, saying users could continue with LangChain or LlamaIndex if they were satisfied with them.21:40
Reading groupHow AgentOps Enables ObservabilityClaimAn agentic AI system can be understood as having four capabilities: perception, planning and reasoning, action, and adaptation.2:04