From My Desk
From My Desk is where I think through ideas at the intersection of AI, engineering, architecture, security and leadership.
Sometimes an idea becomes an article. Sometimes I build a small Lab to see whether it holds up in practice. The goal is not to present every answer as finished, but to make the thinking visible enough to examine, question and improve.
Start with the latest Lab, or browse the earlier articles, Labs and experiments that led to it.
Latest Lab · Lab 007
Can We Prove the Agent Is Actually Contained?
A stop command is not proof. Test whether runtimes, identities, credentials, delegates, queues, tools, networks, state, and recovery paths are independently verified as contained.
The scenarios are fictional, and the Lab never controls a real agent, credential, process, or network path.
Lab 006
When a valid stop is ignored, containment must stay outside the agent.
Lab 005
A useful receipt binds intent, authority, judgment, execution, and outcome—not only an activity line.
Lab 004
A trustworthy agent is not one that always finishes the task. It recognizes when continuing alone would be unsafe.
Lab 003
What changes when an agent does more than connect and begins acting through tools, data and delegated authority?
Lab 002
What happens when an agent passes part of its authority to another agent? Follow the chain and see why authority may stay within the same boundary or become narrower at each handoff, but must never expand.
Lab 001
Identity tells us which agent is acting. The first Lab explores why authority, limits and context still have to be evaluated before an action proceeds.
I publish the longer essays on LinkedIn. The Labs here are where I test some of those ideas and make the architecture easier to walk through.
