Oak — Git replacement designed for agents. Oak reimagines Git as an agent-native VCS, enabling agents to share, coordinate, and manage code and state. If you run multi-agent workflows, an agent-aware version control solves coordination, provenance, and merge conflicts at the tool layer.

Lithuanian startup Superpal raises €500K for AI coworker platform built inside Slack. Superpal launches an autonomous AI coworker inside Slack that connects to 1,000+ tools, handles end-to-end tasks, and enforces company-level privacy. It’s a practical blueprint for embedding agentic assistants into existing comms and access-control boundaries, showing how orchestration and governance meet in production.

Researchers introduce Self-Harness, a framework that lets AI agents rewrite their own rules, boosting performance up to 60%. Self-Harness lets LLM agents autonomously rewrite their operating rules, improving harnesses by up to 60% through trace-driven edits and regression testing. That capability speeds agent iteration but increases the need for regression suites, audit logs, and immutable harness snapshots so you can validate and roll back self-modifications.

How Claude Mythos found a 15-year-old bug in Mozilla Firefox. Mozilla used an agentic harness plus model scoring and verifier subagents to find and fix a 15-year-old Firefox security bug. It’s a concrete example of agent-assisted discovery paired with verifier subagents and human triage—exactly the operational pattern you’ll need for safe, auditable outcome engineering.

Agentics: Cost to Implement vs Cost to Verify. The piece argues that verification costs, not model throughput, decide when delegating code to agents is safe. For outcome engineers that means investing first in verifiers, test harnesses, and audit tooling—otherwise scaling delegation multiplies risk, not value.