Governing Persistent AI Agents (opens in a new tab)
Helixar Research
A technical governance model for agents that persist across sessions, focusing on identity, delegated authority, memory integrity, ownership and lifecycle management.
Issue #003 · September 6, 2026
AI DevList #003
Multi-agent systems, MCP, agent skills, observability and production AI.
A focused selection of what mattered this week for engineers building AI systems.
Agentic AI
AI Security & Red Teaming
🔥 Inside 90 Days of Attacks on AI Infrastructure (opens in a new tab)
Wiz Threat Research
Wiz deployed honeypots across LiteLLM, Flowise, LangChain, Langflow, ChromaDB, Ollama and other AI infrastructure, observing real attacks involving MCP remote-code execution, blind prompt injection and AI-specific post-exploitation.
Anatomy of a Frontier Lab Agent Intrusion (opens in a new tab)
Hugging Face security team
A detailed reconstruction of a multi-day agent-driven intrusion involving lateral movement, credential exposure, trust-boundary failures and thousands of autonomous actions.
Tools & Frameworks
DeepSeek Harness: Why the Agent Layer Is Becoming Replaceable (opens in a new tab)
Beagle
A technical breakdown of DeepSeek Harness and its plugin-first design, where models, tools and orchestration components can be swapped independently.
Research & Papers
EnergyEvals (opens in a new tab)
Tume AI
A tool-augmented agent benchmark covering real-world energy-market retrieval, regulatory interpretation and quantitative analysis.
Production & Evaluation
Infer-forge: Harness, Loop and Graph Engineering Around SGLang (opens in a new tab)
LMSYS / SGLang community
Infer-forge explores how agents can perform long-running systems and inference-engineering work while maintaining reproducibility, verification and explicit task graphs.
AI Evals Are Becoming the New Compute Bottleneck (opens in a new tab)
Hugging Face community
A data-heavy analysis of how agent evaluations are becoming expensive enough to form their own infrastructure budget, particularly when teams measure consistency rather than a single successful run.
Open Source
Hermes Agent v0.20.6 (opens in a new tab)
Open Source / Agents
The latest Hermes Agent release rolls up hundreds of merged pull requests spanning browser control, remote MCP integrations, secret storage, terminal backends, caching and long-running agent operations.
Tutorials & Deep Dives
DeepSeek Harness Isn't a Claude Code Killer. It's Something More Interesting. (opens in a new tab)
Harness the Agents
A hands-on examination of the DeepSeek Harness architecture, including its plugin model, permission modes, session logs and ability to delegate to other coding agents.
Enjoyed this issue?
Get the next one straight to your inbox.