PodcastMachine Learning, AI Agents, and AutonomyPushed backEgor Kraev disputes the idea that LLMs should be judged as complete solutions or agents, arguing that they are additional components in a larger system.9:33
Threads / Agents in production / 2025
Agents in production in 2025
94 sessions
PodcastMachine Learning, AI Agents, and Autonomy
PodcastReal World AI Agent StoriesClaimA reliable agent system depends on having quality user, system, and architecture data available to the models.10:56
PodcastBuilding a Decentralized Knowledge Graph for AIClaimEdge nodes act as lightweight gateways around an organization’s or agent’s own Knowledge Graph, allowing systems to publish and access data without duplicating it.10:34
PodcastAutonomous AI SRE: The Future of Site Reliability EngineeringClaimWillem Pienaar says production operations are harder for AI agents than development because production lacks complete datasets of past problems and solutions.3:10
Reading groupAI Agents: The Future of ML Engineering?Pushed backSuccess on Kaggle competitions was disputed as evidence that an agent can perform general machine learning engineering or automate scientific discovery.7:38
Building AI That Remembers YouClaimLetta is an open-source framework and platform for building persistent agents with reasoning, memory, and personalization.0:00
PodcastThe Agent Landscape - Lessons Learned Putting Agents Into ProductionPushed backThe speakers reject the idea that declining cost per token automatically means that agent systems are becoming cheaper overall, because cost per answer can rise as agents use more tokens and calls.21:33
AI Agents: The Evolution of Our Digital FriendsClaimAn LLM is like the reasoning engine of a Raspberry Pi, while an agentic system connects that engine to memory, tools, and the world.0:40
PodcastWeb Agents: The Cutting Edge of AI is Here?Pushed backThe team found that performance on the WebArena benchmark did not translate reliably to the web-agent tasks they cared about.17:49
The Future of AI: Building Multi-Agent SystemsClaimAdding enterprise-specific constraints can reduce an agent's effectiveness, especially when the business knowledge was not represented in its training data.1:18
PodcastAI SQL Data AnalystPushed backThe speakers rejected the view that the agent should replace data analysts, emphasizing that it should handle first-line questions while humans handle complex analysis.42:05
PodcastThe Agent Exchange: Practitioner InsightsPushed backDmitri Jarnikov disputes the assumption that platforms can simply block generic agents, arguing that blocking can make the experience worse for real users and will be difficult to sustain.32:52
PodcastFuture of Software, Agents in the Enterprise, and Inception Stage Company BuildingClaimEliot Durbin says CrewAI improved agent accuracy by separating agents into crews and assigning them discrete roles and tasks.10:27
PodcastI Let An AI Play Pokémon! - Claude plays Pokémon CreatorClaimDavid Hershey created Claude Plays Pokémon as a side project to build and learn about AI agents.2:00
PodcastUnbearable Lightness of Data: The Weight of InformationPushed backRohit rejected the idea that AI agents can exist without a physical substrate, arguing that they require hardware such as H100 GPUs and corresponding power and replacement capacity.32:18
PodcastFrom Rules to Reasoning EnginesPushed backGeorge Mathew disputes the idea that the future will mainly consist of copilots assisting humans, arguing that fully autonomous autopilots will also emerge.18:12
PodcastBeyond the Matrix: AI and the Future of Human CreativityClaimFausto Albers proposes using AI interviews and virtual agents to understand community members and recommend useful connections and activities.16:16
PodcastFraud Detection in the AI EraClaimRafael Sandroni says AI assistants need testing for response quality, orchestration, API integrations, and regression prevention.1:40
Build a Slack Agent with Pydantic AIClaimThe tutorial builds a Slack Digest agent that finds important Slack threads and summarizes them.0:23
PodcastAI Data Engineers: Data Engineering After AIPushed backVikram Chennai rejected the idea that full agent autonomy is necessarily the right endpoint and argued that users should retain control over production changes.25:29
Reading groupA-MEM: Agentic Memory for LLM AgentsPushed backA participant questions why brute-force inclusion of a 9,000-token conversation history in a 128k context window did not always outperform memory lookup.55:56
PodcastBehavior Modeling, Secondary AI Effects, Bias Reduction & Synthetic DataPushed backDevansh Devansh disputed the common emphasis on autonomous agents by arguing that they are generally unsuitable for business use and that deterministic workflows work better.38:16
Iceberg, MCP, and MLOps: Bridging the Gaps for EnterprisePushed backHamza Tahir rejects enforcing one fixed MLOps stack for every enterprise context.23:37
AI Agents for Financial InstitutionsPushed backA functional agent that authenticates users, retrieves context, and performs actions still needs an additional validation and reasoning stage before production use.8:11
Evaluation of Agentic SystemPushed backAditya Gautam distinguished holistic stress testing of a complete agentic system from isolated testing of individual APIs, LLMs, or agents.22:54
Building an AI agent with LangGraph, step by step tutorialClaimAn agent is appropriate for problems where AI roles make decisions at different points in a framework, even when the agent also uses retrieval-augmented generation or chatbots.0:23
PodcastHow AI is Revolutionizing Marketing ForeverPushed backFausto disagreed with the claim that AI-generated marketing is over, arguing that marketing will continue but its methods will change as AI agents become decision makers.10:46
MCP is not going to change everything (yet)Pushed backA2A was presented as potentially important for business delegation, but Sam Partee questioned whether it was needed because ordinary tool calls can already implement observable agent handoffs.16:05
Reading groupMem0: Building Production-Ready AI Agents with Scalable Long-Term MemoryClaimPrateek Chhikara says current LLMs are good at short-term coherence but forget information across sessions because context windows are fixed.4:57
Everything Hard About Building AI Agents TodayPushed backThey distinguish synchronous interactive agents from asynchronous background agents and disagree with treating their feedback and control requirements as the same.22:25
PodcastHow AI is Transforming Business Data ForeverPushed backDeepti Srivastava disputes the idea that Model Context Protocol by itself solves the connector problem.36:27
Building AI that Doesn't BreakPushed backAlan Nichol disputes the view that atomic function or tool calling is the right abstraction for customer-facing agents.38:42
A New Way of Building with AIPushed backCode-based orchestration is more suitable than repeated function-calling for large repetitive jobs because it is less expensive and scales across rows.27:18
PodcastFrom the Legal Trenches to TechClaimLexMed provides tools that use large language models to streamline the work of Social Security disability attorneys.0:56
AI Agent Development Tradeoffs You NEED to KnowPushed backLangGraph's graph and workflow structure was presented as potentially more complex than necessary compared with simpler agent loops.10:21
MLflow 3.0: The Future of AI AgentsClaimMLflow added significant capabilities for building GenAI applications and agents around 2024.0:38
PodcastWhy Enterprise AI is Taking So Long to ImplementPushed backDemetrios Brinkmann suggests that broad company context may be important for agents, while Sean Kenny argues that the specific context needed to complete a job is more important than general company context.1:00:55
9 Commandments Building AI AgentsPushed backDemetrios Brinkmann treated humans as tools, while Dmitri Jarnikov initially said he had been thinking about other agents rather than humans as tools.46:47
AI Coding Agents Change Software Development ForeverPushed backThe command line alone is enough for coding agents.1:51
The Hidden Bottlenecks Slowing Down AI AgentsPushed backThe speakers reject the idea that the evaluation tool itself is normally the main evaluation bottleneck.1:04
Prepare Your Data for AI Now, or You're Screwed LaterPushed backDirectly converting enterprise APIs into MCP tools does not work beyond trivial use cases.7:08
Advanced Context EngineeringPushed backDex rejects the idea that tool use should be framed as an alien entity interacting with the world, saying it is ordinary software that produces JSON and executes deterministic code.9:54
How Synthetic Data Will Transform Publishing ForeverClaimAsk News is intended to help sustain publishers by providing grounded synthetic data for AI agents.0:00
The Science of Improving AI AgentsClaimEval-Driven Development applies the scientific method to improving AI agents by forming hypotheses, testing them, validating the results, and acting on the findings.1:34
When Agents Hire Their Own Team: Inside Hypermode's ConciergePushed backThe moderator asks Ryan Fox-Tyler to reconcile independent agents that communicate by message passing with the industry view that agents should continuously contribute to shared state.19:57
PodcastKnowledge is Eventually ConsistentPushed backDemetrios Brinkmann argues that an agent’s unsolicited long answer can add cognitive load instead of value.12:30
AI Changed Stack Overflow for the BetterClaimStack Overflow's detailed, evolving answers help with questions that AI systems cannot answer because they are highly specific or context-dependent.5:19
How AI Will Transform The Energy SectorClaimAdam Sroka says Hypercube is launching Jellyfish, an agentic AI product for large energy infrastructure projects.2:03
I Built A Trustworthy Voice AssistantClaimA voice agent listens, understands, decides, and responds in natural language, turning human speech into structured, actionable interactions.0:48
How Agents Changed Vibe Coding ForeverPushed backBeyang Liu argues that the IDE, particularly the AI IDE or VS Code fork, is outdated as the central coding interface for the agentic era.3:05
APICA: The Digital Colleague at the Port of Antwerp-BrugesClaimAPICA Chat is intended to become the interface for the port's entire smart system and to route questions to the appropriate specialized agent.2:36
Deploying Executable Agent WorkflowsPushed backGal Peretz argued that code generation can address workflow limitations that are difficult to express with JSON tool calling.10:59
AI traces are worth a thousand logsPushed backTraces should not be treated as part of the agent's context; they are a separate view of the operations the agent performs.10:48
PodcastDistilling 200+ Hours of NeurIPS: What's Next for AIPushed backNikolaos Vasiloglou argued that agents should currently support decisions rather than make and execute important decisions autonomously.21:41
From Spikes to Stories: AI-Augmented Troubleshooting in the Network WildClaimThe main troubleshooting pain points are cognitive overload, loss of context during investigation, and difficulty interpreting network operations data without specialist expertise.3:08
The Hidden Infrastructure Behind Every AI AgentClaimEvery time an AI agent takes an action, queries a model, calls an API, or fetches a tool, something has to handle that traffic.1:21
Too much lock-in for too little gain: agent frameworks are a dead-endPushed backSimple composable patterns and complex agent frameworks address different needs, so builders should balance reuse and production capabilities rather than choose only one approach.4:59
The Future of User Interfaces: AI is Changing EverythingPushed backTools should not guess that a similarly named user was intended when the exact requested username is not found.16:40
Advancing the Cost-Quality Frontier in Agentic AIClaimKrista Opsahl-Ong works on productionizing agents for enterprises at Databricks and on improving the cost-quality frontier.0:00
Cutting Costs with Artificial IntelligenceClaimMulti-agent systems make cost tracking harder because agents communicate with one another and with users while input and output sizes vary.0:52
PodcastThe Era of AI Agents in MarketingPushed backAI agents will eliminate the need for sales development representatives.9:35
The Future of Compute: How AI Agents Are Reshaping InfrastructurePushed backDiego Oppenheimer disputes the assumption that existing infrastructure paradigms will necessarily be sufficient for agentic workloads, while also rejecting an automatic call to reinvent everything.28:07
Zero Trust for Multi-Agent SystemsClaimA multi-agent system is a network of independent autonomous software agents designed to perform specific tasks based on assigned roles.1:19
Before Building AI Agents Watch These Hard Earned LessonsPushed backNishikant Dhanuka disputes the view that the latest model or semantic search alone solves an agent's problem, arguing that context and additional query-processing steps are needed.5:20
AI Needs Memory: Here's How It WorksPushed backThe speaker argues that human-style memory categories are flawed as a literal description of agent memory but still useful as a working model.7:18
Building Multi-Player AI Systems (and why it's SO hard)ClaimMeshAgent focuses on enabling an entire team, rather than only one person, to interact with agents.0:58
Reading groupBeyond Prompting: The Emerging Discipline of Context Engineering Reading GroupPushed backThe speakers distinguish context engineering from prompt engineering rather than treating them as the same activity.5:25
Building Agents for HealthcareClaimCorti combines a healthcare foundation model with patient information, guidelines, workflows, and tools to support healthcare use cases.5:58
Designing AI Agents for the Complex Realities of HealthcarePushed backPhysicians may face liability whether they follow an AI-agent recommendation that is wrong or reject one that would have been right.7:57
The Cost of AI: FinOps Strategies for Intelligent AgentsClaimAdvait Patel says autonomous AI agents are not necessarily optimized and can increase cloud costs when deployed incorrectly.2:26
How to Optimize AI Agents in ProductionPushed backNimrod disputes the idea that agent improvement can be handled by testing one configuration at a time or by relying only on A/B testing.12:40
What It Takes to Run Multi-Agent SystemsClaimMulti-agent workflows can consume millions of tokens per session, making a fully cloud-based AI stack expensive.3:58
Beyond Chatbots: How to build Agentic AI systems with Google GeminiPushed backAgent evaluation should focus on reliability across repeated runs, not only capability or correctness on a single run.13:14
How to Build Execution Layers That Don't Burn OutClaimAgentic systems can lose their main intention when they act on individual inputs, especially when orchestrating multiple APIs.0:51
Underwriting Assist: A Multi-Agent SystemClaimPalona AI builds fully autonomous agents for restaurants, serving guests and helping restaurant operators improve consistency and efficiency.0:00
How to build agents that take ACTIONPushed backThe audience member argues that data quality is foundational, while Alex Salazar argues that it is less important as a first step for current agent development.24:48
Overcoming Agentic Memory Management ChallengesPushed backWhether AI agents need to forget remains disputed because some people favor pruning outdated information while others favor retaining and archiving information without making agents behave like humans.10:44
Fine-Tuned Models Are Getting Out of HandPushed backJaipal argued that agents should not usually be given complete autonomy in creative, information-sparse workflows.27:59
PodcastThe Claude Code Story: How Two Guys Changed AI ForeverClaimClaude Code began as a Labs team prototype that accessed Claude from a terminal without code-writing or tool capabilities.0:49
Meta-Prompting: The Hack That's Changing Production AIClaimArize is an engineering platform for people building AI agents that helps them evaluate and observe those agents in production.0:05
AI Agents Are About to Get Real, Here's Why It MattersPushed backModel Context Protocol is commonly treated as tool calling, but Sam Partee says it primarily gathers context and does not itself execute or mutate external resources.2:41
PodcastThe Real Reason Your AI Models ForgetClaimJeff Huber says models become less able to pay attention and reason as more context is added.1:56
The Next Evolution of AI AgentsClaimAI agents in production face three major challenges: confidence, security, and efficiency.5:07
Sub-Agent Architectures: What You Can LeveragePushed backDrag-and-drop agent-building tools are no longer the right approach.1:33
The Vector Search Engine Powering AI AgentsPushed backThe first AI agent was not created in 2020; Thierry Damiba placed its origin around the 1960s with ELIZA.0:43
MCP-Enabled AgentsPushed backSophia Skowronski questioned whether adding many distractor MCP servers truly replicates production conditions, rather than merely making the benchmark harder.27:08
Linear's Approach to Production AI AgentsClaimLinear is a project management tool for product teams that provides a shared workspace for engineering, product, and go-to-market teams.0:29
PodcastVoice AI's Biggest Weakness ExposedPushed backBrooke Hopkins says user impatience, rather than losing conversation context, is the main immediate consequence of a voice-system delay.19:40
PodcastGovernance for AI Agent DeploymentPushed backSpencer Reagan argues that retrieval-augmented generation based on matching chunks will diminish in applicability as agents gain the ability to retrieve information through tools.48:48
PodcastHow Sierra AI Does Context EngineeringPushed backZack argued that goals and guardrails describe Sierra's agents better than traditional decision trees, while still treating decision trees as potentially useful concepts.47:59
PodcastDoes AgenticRAG Really Work?Pushed backDemetrios Brinkmann suggests that an MCP server for a database might replace or simplify the agent approach, while Satish Bhambri says the choice depends on the use case and the resources and information that need to be exposed.13:37
Big Updates to Redis: Context Engine, Tools, and MemoryPushed backSimba Khadder disputes the assumption that direct text-to-SQL is an acceptable way to let agents access production databases.7:10
PodcastContext Engineering 2.0Pushed backContext engineering should not be reduced to conventional RAG, knowledge graphs, or embedding every piece of structured data.14:29
PodcastThe Future of AI Agents are SandboxesPushed backThe host suggested that agents might need whole databases copied into sandboxes, while Jonathan Wall explained that scoped read-only MCP credentials, object storage, database clones, or other approaches can also provide safe access.12:53