Agent Ops: Mesh LLM, agent traces, and supply-chain alarms

Mesh LLM: distributed AI computing on iroh launches a GPU mesh using iroh that serves an OpenAI-compatible API and shards models across machines. This gives outcome engineers a practical horizontal inference fabric for agent fleets, lowering provider lock-in and forcing new deployment and orchestration patterns (Principles 01,09).

Mindwalk — Replay coding-agent sessions on a 3D map of your codebase visualizes coding-agent sessions as glowing paths on a 3D city-map of your repo so you can replay what the agent actually explored and edited. That provides observable agent traces for debugging, audit, and documentation—essential for legible landscapes and accountable agents (Principles 06,13).

What xAI’s Grok Build CLI Actually Sends to xAI — Wire-Level Analysis demonstrates the Grok CLI uploading entire repositories and unredacted secrets to xAI’s storage, backed by reproducible wire captures. Outcome engineers must treat developer tooling as a high-risk exfiltration vector and add client-side scanning, provenance controls, and stronger Gate and Immune System protections (Principles 02,14).

Forget typosquatting; slopsquatting is the software supply chain threat created by AI coding tools documents “slopsquatting,” where attackers register fabricated package names suggested by AI assistants and inject malware into real projects. This reframes dependency hygiene: lock registries, validate provenance in CI/CD, and bake supply-chain checks into your delivery pipeline to prevent agent-driven supply-chain compromises (Principles 02,14).

Fixed three bugs that made Qwen3.5-122B a daily driver on Mac Studio fixes serving-stack bugs to run Qwen 3.5 122B reliably as a local daily driver on an M3 Mac Studio. Local heavy-model serving becomes viable for latency, privacy, and offline agents—outcome engineers should plan for local context caching, model lifecycle management, and hybrid deployment topologies (Principles 07,11).