エピソード

  • The Agentic Engineer Podcast - Issue #21: GPT-5.6 Goes GA — Programmatic Tool Calling Changes Everything

    The Agentic Engineer Podcast - Issue #21: GPT-5.6 Goes GA — Programmatic Tool Calling Changes Everything
    2026/07/15
    GPT-5.6 ships three tiers and Programmatic Tool Calling. Anthropic discovers Claude's silent internal workspace (J-Space). CubeSandbox gives agents hardware-isolated sandboxes in 60ms. Plus GRAM, OfficeCLI, Loom for AWS, and why Anthropic is an unreliable narrator.
    続きを読む 一部表示
    13 分
  • The Agentic Engineer Podcast - Issue #20: Better Models, Worse Tools

    The Agentic Engineer Podcast - Issue #20: Better Models, Worse Tools
    2026/07/08
    Opus 4.8 and Sonnet 5 invent fake schema fields when calling third-party tools. Armin Ronacher investigated why RL training on Claude Code creates an accidental moat. OpenAI Codex accounts for 99.8% of all tokens internally. Strix hits #1 trending for AI pentesting. AWS CloudFormation Express Mode ships 4x faster deploys. Code cleanliness paper shows 7-8% token savings. And Zuckerberg says agents are hard while OpenAI says they work perfectly.
    続きを読む 一部表示
    10 分
  • The Agentic Engineer Podcast - Issue #19: GPT-5.6 Sol — The First Government-Gated Model Release

    The Agentic Engineer Podcast - Issue #19: GPT-5.6 Sol — The First Government-Gated Model Release
    2026/07/01
    OpenAI gates GPT-5.6 Sol behind government pre-approval. GLM 5.2 beats Claude Opus 4.8 on Semgrep's IDOR benchmark at 1/6 the cost. Google ships DESIGN.md for coding agent visual identity. AWS Lambda MicroVMs launch as GA serverless sandboxes for AI agents.
    続きを読む 一部表示
    11 分
  • The Agentic Engineer Podcast - Issue #18: Claude Requires Government ID — The Trust Fracture

    The Agentic Engineer Podcast - Issue #18: Claude Requires Government ID — The Trust Fracture
    2026/06/24
    Anthropic requires government ID for Claude access. AWS Summit NYC expands AgentCore with managed knowledge base and web search. codebase-memory-mcp gains 6,372 stars. Flue ships sandbox agent framework for TypeScript. Project Fetch Phase 2 does robotics 20x faster. AWS Continuum launches autonomous security at machine speed.
    続きを読む 一部表示
    10 分
  • The Agentic Engineer Podcast - Issue #17: The First Frontier Model Recall

    The Agentic Engineer Podcast - Issue #17: The First Frontier Model Recall
    2026/06/17
    US government pulls Fable 5 and Mythos 5 from production. AWS ships three agentic services: DevOps Agent, FinOps Agent, OpenSearch MCP Apps. NVIDIA SkillSpector scans agent skills for 64 vulnerability patterns. AWS Blocks launches Infrastructure from Code.
    続きを読む 一部表示
    14 分
  • The Agentic Engineer Podcast - Issue #16: Agentic Commerce Goes B2B

    The Agentic Engineer Podcast - Issue #16: Agentic Commerce Goes B2B
    2026/06/10
    Amazon licenses its agentic shopping assistant to outside retailers. Perplexity publishes production data showing autonomous agents cut knowledge work time 87%. DeepSeek V4 Pro beats GPT-5.5 Pro on precision. Claude Opus 4.8 and Fable 5 land on AWS Bedrock. Ladybird kills public PRs.
    続きを読む 一部表示
    10 分
  • The Agentic Engineer Podcast - Issue #15: The Plugin Wars Begin

    The Agentic Engineer Podcast - Issue #15: The Plugin Wars Begin
    2026/06/03
    Anthropic open-sources 11 knowledge-work plugins for Claude Cowork. OpenSearch Serverless Next-Gen kills the $300/mo minimum with true scale-to-zero. Self-improving agents go from 25% to 86% accuracy. Human-in-the-loop is a liability when users approve 93% without reading.
    続きを読む 一部表示
    17 分
  • The Agentic Engineer Podcast - Issue #14: Mythos 10K Vulns, CodeGraph, Constraint Decay

    The Agentic Engineer Podcast - Issue #14: Mythos 10K Vulns, CodeGraph, Constraint Decay
    2026/05/27
    Anthropic's Mythos Preview found 10,000+ critical vulnerabilities in 30 days. A new paper proves coding agents collapse in convention-heavy frameworks like Django. CodeGraph cuts agent costs 35% with pre-indexed knowledge graphs.
    続きを読む 一部表示
    18 分