AgentStack Daily

The Agent Stack Release Readout this week leads with Hermes Agent v2026.7.1 shipping alongside Claude Code CLI 2.1.193, with Kimi K2.7 Code reaching GA inside GitHub Copilot. ZCode wraps GLM-5.2 and lands on the Hacker News front page, Leanstral 1.5 brings open-weights Lean proof generation, and Alibaba bars Claude Code at work over backdoor concerns. Jamesob's GitHub guide curates locally-runnable SOTA LLMs, WebBrain ships a local-first browser agent for Chrome and Firefox, ReContext publishes a training-free long-context harness, and Snorkel releases Senior SWE-Bench. Plus ghealth CLI, Program-as-Weights, DramaSR-532K, AgenticSTS, and the Safari MCP server.

Show notes: https://tobyonfitnesstech.com/podcasts/episode-79/

Agent Stack Daily covers the return of Claude Fable 5: US restrictions on Anthropic's Mythos and Fable models were dropped on June 30, and the Mythos-class flagship is back on OpenRouter with 1M context, multimodal input, and reasoning support. OpenClaw v2026.6.11 and OpenAI Codex rust-v0.142.5 shipped, and Claude Sonnet 5 hit OpenRouter with 1M context alongside Google's Nano Banana 2 Lite. Research includes the Orca unified world latent space paper, the Agents-A1 35B MoE, BlockPilot on diffusion speculative decoding, generative skill composition, and TRIAGE on role-typed credit assignment. OmniRoute consolidates 231 providers behind one endpoint.

Show notes: https://tobyonfitnesstech.com/podcasts/episode-78/

Today's AgentStack Daily covers Claude Code CLI 2.1.185, Cursor's new mobile app for steering coding agents, and Google making Gemini's personalized image generation free for US users. Also: OKX's plan to let AI agents hire and pay each other on crypto rails, Base44's custom model release, and Anthropic's California government Claude pricing cut. Plus South Korea's $550B memory fab bet, the AI jobs debate, a founder's cancer fight using AI, and an AINews segment with little to report.

Show notes: https://tobyonfitnesstech.com/podcasts/episode-77/

Today's AgentStack Daily covers OpenAI Codex rust-v0.142.4 shipping as a chores-only maintenance release, HP Inc. launching a Frontier strategic partnership with OpenAI across engineering and the 100,000+ partner portal, and OpenAI + Broadcom unveiling the Jalapeño inference chip. We track a founder's Claude-based personal cancer treatment coordinator, Ford rehiring senior engineers after AI quality push stumbles, and Micron becoming Wall Street's next AI memory bet on HBM. We close with SoftBank and Sam Altman questioning Musk's orbital data center pitch and Apple Vision Pro VP Paul Meade reportedly joining OpenAI's hardware team.

Show notes: https://tobyonfitnesstech.com/podcasts/episode-76/

Today's AgentStack Daily covers OpenAI's GPT-5.6 Sol preview for coding and science, and the White House asking OpenAI to slow-roll the launch. Tooling updates: OpenClaw v2026.6.10, OpenAI Codex rust-v0.142.3, and Claude Code CLI 2.1.181. Patronus AI raises $50M for agent stress-test 'digital worlds.' OpenAI reports median Codex output tokens grew 56x in Research, 32x in Customer Support. Databricks' Matei Zaharia and Reynold Xin argue the frontier ecosystem must stay open as OpenAI, SpaceX, and others build custom chips, with OpenAI's Jalapeño turning up heat on Nvidia. New data suggests engineering jobs are the most AI-resilient.

Show notes: https://tobyonfitnesstech.com/podcasts/episode-75/

OpenAI Codex 0.142 is the new stable, with usage-limit reset credits, organized plugins, and rollout token budgets. OpenAI's Daybreak initiative brings Codex Security and GPT-5.5-Cyber for vulnerability discovery, plus Patch the Planet for open-source maintainers. Samsung is rolling out ChatGPT Enterprise and Codex globally, and Nex AGI listed Nex-N2-Pro on OpenRouter. sqlite-utils ships migrations and nested transactions in 4.0rc1, iOS 27 brings practical AI features below the Siri surface, SpaceX inked a $150M/month compute deal with Reflection AI, and Groq confirmed a $650M raise. Also: the agentic "loopy" pattern, Fika Jobs' AI interview agents, and Z.ai's GLM-5V Turbo multimodal agent model.

Show notes: https://tobyonfitnesstech.com/podcasts/episode-74/

Today's AgentStack Daily: OpenClaw v2026.6.9, Hermes Agent v2026.6.19, and Claude Code CLI 2.1.176 all shipped new releases. Poolside released Laguna XS.2 on OpenRouter and Laguna M.1 via API. Enterprise teams got new usage analytics and updated spend controls. A retrospective asks whether 30 years of export controls can contain a model called Mythos. Baseten is reportedly raising $1.5B at a $13B valuation. Datasette Apps launched for hosting custom HTML inside Datasette. Meredith Whittaker of Signal says AI chatbots are not your friends. In the Weights debuts as a vanity search engine for AI.

Show notes: https://tobyonfitnesstech.com/podcasts/episode-73/

This episode covers OpenClaw v2026.6.8, OpenAI Codex rust-v0.141.0, and Claude Code CLI 2.1.170. Google's Nano Banana 2 and Nano Banana Pro image models are listed on OpenRouter. OpenAI rolls out LifeSciBench and a pre-release deployment simulation, and teams with Molecule.one on a GPT-5.4 medicinal chemistry reaction. Z.ai releases GLM-5.2 open weights under MIT, claiming a top open-weights slot via IndexShare speculative decoding. Radical AI argues the lab is the moat, and NEA's Tiffany Luck weighs in on enterprise AI ROI.

Show notes: https://tobyonfitnesstech.com/podcasts/episode-72/

Today's episode covers the latest developments in AI, including the stable release of OpenAI's Codex rust-v0.140.0, new foundation models from Apple, and significant acquisitions, such as SpaceX's $60B purchase of Cursor. Tune in for analysis on the rapidly shifting AI industry, including major investments, layoffs, and the launch of new AI agent identities by NewCore.

Show notes: https://tobyonfitnesstech.com/podcasts/episode-71/

A heterogeneous dual-GPU setup running an RTX 5080 paired with an RTX 3090 sustained 80 tokens per second on Qwen 3.6 27B at Q8 quantization with a clean tensor-parallel layer split. OpenAI opened Codex to open source maintainers and added three workflow courses in OpenAI Academy. Anthropic suspended Fable 5 and Mythos 5 from US government access following a federal directive tied to the Amazon CEO's meetings with Trump administration officials. Endor Labs placed Fable 5's coding results mid-tier, and developers debated Claude Fable's proactive agent behavior.

Show notes: https://tobyonfitnesstech.com/podcasts/episode-70/