Now

What does it take to stay visible, measurable, and operational when AI systems mediate discovery and work?

Two tracks, really: how people and businesses get found when answers replace clicks, and how agent workflows keep context, recover from failure, and stay inspectable when they touch real tools.

• Marketing operations agents at Bobdo

  Long-running agents around campaign monitoring, SEO audits, analytics, and report generation across multiple clients. The interesting part isn’t the automation. It’s coordination: watching systems that change, preserving context, and handing work back to humans at the right moment.

• Beetroot

  Continuing work on a local-first clipboard environment for Windows. How much workflow context actually passes through the clipboard, and what changes when that context becomes searchable and durable?

• max.nardit.com

  Restructuring this site around a clearer systems identity.

• Long-running agents

  What breaks when an agent runs for days or weeks instead of a single chat session: restarts, drift, stale assumptions, and invisible state.

• Operational memory

  How agents should remember across sessions without turning memory into an unbounded junk drawer.

• Orchestration

  How to coordinate multiple agents on overlapping work: task routing, shared state, and human approval.

• Tool boundaries

  Where MCP helps, where it leaks, and what still has to be handled by the surrounding system.

• Human handoff

  Where an agent should stop, ask, or hand control back to a person.

• Agent reliability patterns

  Less “how to make an agent do X” and more “how to know what happened, recover from it, and avoid corrupting state.”

• Japanese

  Active study right now.