← Latest Update

Agents, Memory, and Safety: 5 Practical Updates for Outcome Engineers

Groundcover expands Agent Mode with Slack, Linear and GitHub connectors — Groundcover now lets AI agents act directly in Slack, Linear and GitHub to recommend code, open PRs, and manage tasks from observability data. This turns agentic suggestions into executable developer work and forces teams to design safe, auditable action lanes for agents (Principles 03 & 09).

Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding — DeepReinforce releases Ornith-1.0, an open-weight LLM family built for self-scaffolding agentic coding workflows that run locally with MoE and dense variants. Running capable, open models on-premises lets you build islanded agent harnesses and lowers dependency on hosted stacks, changing how you validate and govern agent behavior (Principles 07 & 09).

Memora: A Harmonic Memory Representation Balancing Abstraction and Specificity — Microsoft Research introduces Memora, a memory design that stores rich content separately from lightweight retrieval abstractions and cuts context tokens by up to 98%. That reduction materially lowers token costs and makes long-horizon agents practical, so rework your retrieval and evaluation pipelines around compact abstractions (Principle 06).

The attack that hijacked Claude Code came through Sentry. Datadog, PagerDuty, and Jira have the same exposure. — An attacker injected fake Sentry error events to issue commands that hijacked coding agents like Claude Code and exposed secrets. Treat observability and CI systems as part of your threat model: lock down connectors, require attestable intent chains, and add adaptive monitoring to detect agentjacking (Principles 10 & 14).

CoreWeave debuts ARIA agent to automate AI research in Weights & Biases — CoreWeave ships ARIA to automate experiment analysis inside Weights & Biases, surfacing insights and recommending model improvements across thousands of runs. Use agents like ARIA to operationalize automated evaluation loops and scale eval-driven development, but bake in audit trails and human checkpoints for model changes (Principles 03 & 16).