AI automation

AI Automation & Agent Systems

I build AI agents and automation that take real actions against real infrastructure. Every one of them keeps the judgment calls with a human — automation removes repetitive work, never the decision.

The stack I use

• Claude / Claude Code for the agentic layer — agents that read code and systems state, make changes, and run multi-step work.
• n8n for workflow orchestration and integration glue between services, schedules, and the agent layer.
• A data-infrastructure mindset carried over from working with Big Data platforms (Cloudera CDH/CDP): pipelines should be observable, versioned, and safe to fail loudly.

The rule I don't break: approval gates

Every system I've shipped keeps a human approval step in front of anything irreversible — publishing content, deploying to production, or resolving something flagged as an anomaly. Automation removes repetitive work, not the judgment call. My own internal AI agent operations system has a dedicated review UI for exactly this: agent actions that need a human to look at them before they're considered resolved. See Projects for the systems this shows up in — Traffic OS's publishing pipeline is the clearest public example of the orchestration in production. For the architecture behind the approval gates and audit trails themselves, see AI Agent Architecture.

What this looks like in practice

• Agents that can propose or take an action, but where anything with real-world consequences routes through an explicit, auditable approval step — the concrete architecture patterns behind that (approval gates as a design pattern, identity checks, immutable artifacts, audit trails) are covered in depth on AI Agent Architecture.

Work with me

If you're evaluating AI agents or automation for your own infrastructure, get in touch via the contact page.