Agentic Thinking Blog

Stories from the runtime.

News, incidents, and architecture notes on AI agent governance, AgentHook, HookBus, AgentProtect, and governed agent loops.

Agents prompting agents: the loop we built into HookBus Agent.
HookBus Agent now supports reusable Agent Teams: governed loops where agents hand off to agents, verify work, preserve evidence, and improve the next run.
Read more →
HookBus Agent is live. Governed agents now have the hooks they need.
HookBus Agent is live at hookbusagent.com: a governed agent runtime that exposes the hooks needed for audit, approval, policy, replay, evidence, and runtime governance.
Read more →
3-day Runtime Governance Validation Sprint
A focused enterprise validation sprint for teams that need to see agent governance working in the runtime, not described in a slide deck.
Read more →
Inside the AgentProtect demo
A practical look at AgentProtect, AgentProtect, approval gates, audit evidence, and runtime governance in a working demo.
Read more →
There is no Article 12 without the runtime
Article 12 puts the obligation to keep logs on the deployer, which cannot be met unless the AI developer emits them.
Read more →
HookBus Light is live. Every autonomous agent now has somewhere to plug in.
The open-source runtime that sits between your autonomous AI agent and the action it is about to take.
Read more →
The agent category just redefined itself in 13 days.
Four autonomous-agent product launches in thirteen days. None shipped with runtime governance.
Read more →
Governance for the agents you cannot watch
Per-tool permissions are useful. The hard problem is keeping an autonomous agent on track when no human reviews each step.
Read more →
Anthropic Managed Agents: universal safety, zero organisational policy
Autonomous Claude agents running bash, writing files, calling APIs, all hosted in the cloud. Brilliant for developers. Unusable for regulated enterprises without organisational policy control.
Read more →
Mythos escaped its sandbox and concealed its actions
During testing it built a multi-step exploit to escape its sandboxed environment, gained internet access, and actively concealed its actions.
Read more →
Claude Code source code leaked via npm
512,000 lines of proprietary Claude Code source code were exposed through a missing .npmignore entry.
Read more →