Exploring Next | Nick Sanders

EP 1121 8m ago Model Behavior Agents Inference Anthropic OpenAI Vercel Agentic Loops State Management In Language Models Tool Use And Function Calling Durable Execution Prompt Injection Context Window Mixture Of Experts Inference Optimization

Model Behavior: Week of October 26, 2026

Anthropic's single-agent push, Vercel's durable workflows, and whether GPT-6.1 Terra and Claude Duet can keep agents coherent when sessions crash.

Search: Tavily Script: GPT-5.5 Voice: Hume Octave 2

EP 1120 8h ago New Models InferenceLaunchOpenAI Gpt 6 1 Terra Claude Opus 5 5Blog

Introducing GPT-6.1 Terra

Delivering fast and capable computer use at scale

Search: Tavily Script: Nemotron 3.5 Super VL 120B A12B Voice: ElevenLabs v4

EP 1119 8h ago Agents Training Trajectory2gym Qwen Supervised Fine Tuning Synthetic Data Generation For Validation Task Decomposition State Management In Language Models Tool Use And Function Calling Agentic Loops Credit AssignmentResearch Paper

Trajectory2Gym: Decision-Grounded Compositional Interaction Synthesis for Tool-Use Agents

Successful tool-use trajectories are not necessarily useful supervision. Even successful interactions may underrepresent multi-step dependencies or contain actions that are unjustified by the information visible to the agent at decision time. We identify two coupled requirements for effective trajectory supervision: compositional coverage, so that important observation-action dependencies are represented, and decision grounding, so that each supervised action is supported by the learner-visible

Search: Bright Data Script: Haiku 4.5 Voice: Inworld TTS 2

EP 1118 8h ago Agents EvalsBenchmarkContinual Learning Exam Claude Opus 5 Agentic Loops Tool Use And Function Calling State Management In Language Models In Context Learning Feedback Loops Long Horizon Training Credit AssignmentResearch Paper

Continual Learning Exam: Benchmarking Long-Horizon RSI and Experiential Learning in LLM Agents

Continual learning is the ability of LLM agents to improve over time by acquiring, refining, and applying knowledge from experience while retaining useful prior knowledge. It is essential for agents learning over long time horizons, yet current evaluations largely assess individual task performance rather than improvement through accumulated experience. We introduce Continual Learning Exam (CLE), a large-scale benchmark developed by 50+ domain experts to measure experience-driven improvement in

Search: SerpAPI Script: Haiku 4.5 Voice: Rime Coda

EP 1117 8h ago Agents Dev ToolsLaunchAnthropic Claude Opus Claude Code Agentic Loops Tool Use And Function Calling Context Window Prompt Engineering Structured Output Model Routing Constraint VerificationBlog

Anthropic says one AI agent beats many subagents now that models can handle complexity alone

Anthropic says companies should evaluate whether they can consolidate sub-agents into a single agent that loads context on demand. In one architecture comparison, the single-agent setup ran about twice as fast. Anthropic has published a migration kit on GitHub to help teams make the transition.

Search: You.com Script: GPT-5.1 Voice: Cartesia TTS

EP 1116 8h ago Agents Dev Tools Dynaflow Qwen Task Decomposition Agentic Loops Tool Use And Function Calling Model Routing Supervised Fine Tuning Fine Tuning On Execution Traces Constraint Verification Directed Acyclic GraphResearch Paper

DynaFlow: Learning Free-Form Multi-Agent Orchestration

Multi-agent systems often rely on predefined collaboration structures, limiting their ability to adapt to diverse task requirements. We introduce DynaFlow, a framework that integrates free-form executable workflow synthesis into the action space of the main agent. The agent decides when collaboration is needed and generates task-specific programs that coordinate sub-agents through sequential or parallel execution, structured information exchange, and verification. Execution results and traces

Search: Jina Script: GPT-5.1 Voice: Hume Octave 2

EP 1115 yesterday Agents Dev Tools Luna 6 Shelley Agentic Loops State Management In Language Models Context Window Tool Use And Function Calling Inference OptimizationBlog

ETOOMANYTHINGS? Run Fewer Agents - exe.dev blog

Spinning up more agents to hide latency wrecks your attention. Fix the human-agent loop instead.

Search: Firecrawl Script: GPT-5.5 Voice: ElevenLabs v4

EP 1114 yesterday Agents Dev Tools Langchain Managed Deep Agents State Management In Language Models Agentic Loops Tool Use And Function Calling Durable Execution Context Window Management Dynamic Code ExecutionBlog

The five components of building a "company harness"

Learn how LangChain built internal agents across marketing, sales, and engineering, and how they have completely changed how those teams work.

Search: Exa Script: Sonnet 4.6 Voice: Rime Coda

EP 1113 4d ago Agents AI Safety Agent Trust Hub Clawhub Aarts Tool Use And Function Calling Agentic Loops Prompt InjectionBlog

You Gave Your AI Agent Access, But Who Is Giving It Orders?

AI agent security risks stem from authority confusion—untrusted content issuing commands to software that already has user permissions—rather than model alignment failures.

Search: Tavily Script: GPT-5.4 Voice: Speechify Simba 3.2

EP 1112 4d ago Inference Dev Tools Llama Cpp Qwen Ollama Mixture Of Experts KV Cache Quantization Sparse Activation Model Routing Inference OptimizationBlog

I changed one llama.cpp flag, and my 8GB laptop GPU now runs a 35B model faster than I can read

8GB of VRAM isn't the bottleneck, your settings probably are

Search: Tavily Script: GPT-5.4 Voice: Inworld TTS 2

EP 1111 4d ago New Models Dev ToolsLaunchOpenAI Gpt 6 Luna Structured Output Token Economics Agentic Loops Tool Use And Function CallingAPI Docs

Build more responsive apps with the Decisions API

Fast, focused, low-cost decisions for any application.

Search: SerpAPI Script: GPT-5.6 Terra Voice: OpenAI TTS

EP 1110 4d ago Agents Dev ToolsLaunchVercel Claude OpenAI Agentic Loops State Management In Language Models Durable Execution Retry Loops And Error Recovery Append Only Logging State Serialization Tool Use And Function CallingTool

Vercel Open-Sources durable-harness for Crash-Proof Coding Agents

Vercel released durable-harness, an open-source package that logs every model and tool call so Pi, OpenCode, Claude Code and Codex sessions resume after crashes, timeouts and redeploys on Vercel, Postgres or a laptop.

Search: You.com Script: GPT-5.4 mini Voice: Cartesia TTS

EP 1109 5d ago Agents Dev ToolsLaunchClaude Duet Anthropic Agentic Loops Tool Use And Function Calling State Management In Language Models Durable Execution Classifier CalibrationResearch Paper

Introducing Claude Duet

Claude Duet is a new app where Claude works as a proactive agent. The research preview opens by waitlist today for Pro and Max subscribers in the US.

No Search Script: GPT-5.4 mini Voice: Hume Octave 2

EP 1108 5d ago Dev Tools OpenAI CodexBlog

What FDE means at OpenAI

What forward deployed engineering means at OpenAI

Search: Firecrawl Script: Haiku 4.5 Voice: ElevenLabs v4

EP 1107 5d ago New Models InferenceLaunchRecursive Flint Mixture Of Experts Quantization Knowledge Distillation Sparse Activation Task Decomposition Context Window Tool Use And Function CallingBlog

Introducing Flint - Recursive

Recursive self-improving superintelligence to automate knowledge discovery.

Search: Exa Script: GPT-5.6 Luna Voice: Rime Coda

EP 1106 5d ago Agents Training Scholarevolve Qwen3 5 27b Gpt 4o Mini Agentic Loops Tool Use And Function Calling State Management In Language Models Retrieval Augmented Generation Task Decomposition Prompt Engineering Fine Tuning On Execution TracesResearch Paper

Learning from Research: Toward Lifelong Agent Harness Evolution

Language agents are expected to solve increasingly complex tasks, creating a growing need for continual improvement. One promising approach is to evolve the agent harness, the software that governs tool use, memory management, and task execution, while keeping the underlying language model fixed. Recent methods automate this process by using a meta coding agent to modify the harness based on execution feedback. However, relying on that agent's existing knowledge and observed failures can

Search: Tavily Script: GPT-5.1 Voice: Inworld TTS 2

EP 1105 5d ago AI Safety Dev Tools OpenAI Gpt 6 Astra Model Routing Anomaly DetectionBlog

Broad access through consistent standards

How OpenAI works with distribution partners to uphold consistent safeguards, support responsible access, and clarify expectations in our Usage Policies.

Search: Bright Data Script: GPT-5.1 Voice: Cartesia TTS

EP 1104 6d ago Agents Agent ObservabilityLaunchCtgt Concordance Qwen Agentic Loops State Management In Language Models Deterministic Inference Prompt Injection Structured Output Model InterpretabilityNews

CTGT releases open source tool Concordance to trace why AI agents make inconsistent decisions

CTGT replaced the identifiers with stable labels, sorted the list and isolated each test in a fresh organizational environment. Five subsequent runs each produced the same 34 obligations.

Search: SerpAPI Script: GPT-5.5 Voice: Cartesia TTS

EP 1103 6d ago Agent Observability Dev ToolsLaunchConcordance Boundaryml Llama 2 Prompt Engineering In Context Learning State Management In Language ModelsResearch Paper

Deterministic Inference Did Not Make Our Pipeline Deterministic

We ran one synthetic banking manual through our pipeline four times and got four different sets of obligations, from a model that returned identical text for identical requests.

Search: You.com Script: Sonnet 4.6 Voice: Hume Octave 2

EP 1102 6d ago Agents Agent Observability Vijil Tool Use And Function Calling State Management In Language Models Agentic Loops Constraint Verification Retry Loops And Error Recovery Append Only LoggingBlog

From feral to fiduciary: making AI agents serve your interests

Vijil argues AI agents should earn autonomy like credit: tested up front, capped by policy, and monitored constantly. Without that, they go feral.

Search: Jina Script: GPT-5.6 Luna Voice: ElevenLabs v4

EP 1101 7d ago Model Behavior New Models Inference Nvidia Nemotron 3 5 Super Vl Reflection Beam Anthropic Mixture Of Experts Fine Tuning Context Window Retrieval Augmented Generation Tool Use And Function Calling Agentic Loops State Management In Language Models Quantization

Model Behavior: Week of October 19, 2026

NVIDIA's Nemotron 3.5 Super VL, Beam's open-weight deployment story, and whether Subquadratic Snap's long-context claims actually reduce agent brittleness—the week's real motion was adoption, not benchmark crowns.

Search: Tavily Script: GPT-5.5 Voice: Speechify Simba 3.2

EP 1100 7d ago Agents Agent Observability Claude Gpt 5 Gemini Agentic Loops Tool Use And Function Calling State Management In Language Models Context Window Management Task Decomposition Retry Loops And Error RecoveryTool

From Microservices to Agents: the Next Evolution of Distributed Systems

The article examines the transition from microservices to agentic AI architecture in distributed systems. Microservices allow for functional decomposition, while agentic systems focus on decision decomposition. This shift introduces new complexities in observability, reliability, and system design, highlighting the need for updated architectural patterns and practices.

Search: Exa Script: GPT-4.1 Voice: Rime Mist v3

EP 1099 7d ago Inference AgentsLaunchLLM D IBM Research Glm 5 2 KV Cache Agentic Loops Context Window Mixture Of Experts State Management In Language Models Inference OptimizationBlog

llm-d 1.0 Advances Open, Cost-Efficient AI Inference for Production Scale

Founded in 2025 by Red Hat, IBM (NYSE: IBM), Google Cloud, CoreWeave and NVIDIA, and now governed by the Cloud Native Computing Foundation (CNCF), llm-d plays a key role in the open foundation for enterprises deploying AI at scale in production.

Search: Tavily Script: GPT-5.4 mini Voice: Fish Audio S2.1 Pro

EP 1098 7d ago Agents Agent Observability Collibra Venturebeat Agentic Loops State Management In Language Models Token Economics Context Window Constraint Verification Append Only Logging State SerializationNews

Scaling AI agents requires inventory, contracts and maintained context

As AI agent fleets grow, enterprises need inventory, enforceable contracts and maintained business context to control risk, cost and reliability.

Search: Tavily Script: Nemotron 3 Ultra 550B A55B Voice: Inworld TTS 1.5 Mini

EP 1097 7d ago New Models InferenceLaunchMistral Timestral Chronos Synthetic Data Generation For Validation Fine Tuning Attention Mechanism Autoregressive Generation Inference Optimization Anomaly Detection Agentic LoopsBlog

Timestral: Mistral's time-series foundation models | Mistral

State-of-the-art probabilistic forecasting in four sizes (5M–209M params), CPU-efficient, zero-shot ready.

Search: SerpAPI Script: GPT-5.1 Voice: Inworld TTS 2

EP 1096 8d ago New Models InferenceLaunchInception Mercury 3 OpenAI Diffusion Models Autoregressive Generation Agentic Loops Tool Use And Function Calling Inference Optimization Context WindowBlog

Diffusion at Scale: Introducing Mercury 3 – Inception

Mercury 3 is the most capable diffusion LLM released to date. It outperforms GPT-6 Luna and Claude Haiku 5.5 while running 4–8x faster at 1,000+ tokens per second.

Search: You.com Script: GPT-5.5 Voice: Hume Octave 2

EP 1095 8d ago Inference Dev ToolsLaunchJev Typesafe AI Databricks Structured Output Calibration Model Routing Entity Resolution Retrieval Augmented Generation Fine TuningNews

Decision models like Jev handle what SQL rules cannot. Here's how data teams use them

Decision models like Jev return typed labels with confidence scores instead of text, positioning them as a cost-effective middle layer between SQL rules and full LLM calls for high-volume classification tasks in data pipelines.

Search: Jina Script: Sonnet 4.6 Voice: Cartesia TTS

EP 1094 8d ago Inference Dev ToolsLaunchNvidia Dory Llama Cpp Ollama KV Cache Quantization Context WindowBlog

NVIDIA Dory 1.0 4B on a Mac: 3.7GB Fits 16GB, Fork Build Needed

NVIDIA's smallest Dory 1.0 4B build is a 3.673 GB Q5_K_M GGUF, so memory is settled on every current Mac. The blocker is software: no llama.cpp release reads the Dory architecture, so you build a pinned contributor fork.

Search: Firecrawl Script: GPT-5.4 Voice: Hume Octave 2

EP 1093 8d ago Agents New ModelsLaunchDeepseek Deepseek V4 1 Flash Claude Opus 5 Mixture Of Experts Transformer Architecture Reinforcement Learning From Human Feedback Retrieval Augmented Generation Tool Use And Function Calling Agentic Loops Token EconomicsAPI Docs

DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API

DeepSeek has just released DeepSeek-V4.1-Flash-1017, a new checkpoint of its V4.1-Flash model. It is live on the DeepSeek website, app and API. DeepSeek claims stronger scores, faster task completion and shorter outputs than the September V4.1-Flash release. The company calls it a Pareto improvement in performance and cost. Alongside it, DeepSeek opened its Search and […]

Search: Exa Script: GPT-5.4 Voice: ElevenLabs v4

EP 1092 8d ago Agents Agent Observability Google Research Harness2 Agentic Loops Tool Use And Function Calling State Management In Language Models Prompt Engineering Task Decomposition Feedback LoopsResearch Paper

Harness^2: Recursive Agent Harnessing for an Open World

LLM agents' capabilities depend not only on their underlying models but also on their execution harnesses. Recursive self-improvement (RSI) offers a promising approach to automatically evolve these harnesses. However, deploying RSI in open-world settings still faces two critical challenges. First, open-world tasks are typically unseen during the harness design phase and novel at test time, leaving no ground-truth verifiers to reliably guide online harness edits. Second, the RSI process remains

Search: Tavily Script: GPT-5.6 Terra Voice: Deepgram Aura-2