Practitioner path
For engineers and ops leads shipping their first or fifth agent. Start with the guides, each of which is a working playbook with code and traces.
education
Education.
I teach four ways: long-form guides for practitioners, opinionated essays for operators, a weekly newsletter for senior marketing leaders, and live cohort workshops where a team builds a working agent system in the room.
All of the written material is free. The cohort and workshop programs are paid and capacity-limited.
manifesto
flagship essay · 11 min read
Responsible AI Needs Architects, Not Just Advocates
Responsible AI has its principles and almost none of the runtime that enforces them. The next credible RAI leaders will ship evals and middleware.
read the essay →learning_paths
Operator path
For senior marketers and AI leaders who decide what gets funded. Subscribe to the newsletter, read the essays, and check the manifesto against your own plans.
Cohort path
For teams that need to ship an agent by a fixed date. The Decoding AI for Marketers cohort is open to the public, and the Think Tank is the closed-door version for senior leaders at member organizations.
canon
The pieces I'd hand to someone joining my team on day one.
// guides
- Claude Code best practices, from a year of daily use How I run Claude Code and the Codex CLI after a year of daily use: MCP profiles, subagent routing, verification rules, hooks, and a starter CLAUDE.md.
- The Content Pipeline: how I run a one-person media operation on n8n How a dozen n8n workflows watch the marketing-AI press, dedupe into one Notion database, post to X and Bluesky, and alert Slack when a run fails.
- Deploying Claude Skills across an organization How to write, distribute, and govern Agent Skills across Claude Code, claude.ai, and ChatGPT, and how to write a description that makes a skill trigger.
- A responsible AI framework for marketing teams The five risks specific to marketing AI, four controls that cover most of them, and a one-week plan to set up governance with free templates.
// essays
- You can't copyright most of what your AI makes, and that's your smaller problem Prompt-only AI output gets no US copyright, and the training-data cases set your liability. What marketers should know as of mid-2026, indemnities included.
- The EU AI Act is a 2026 problem for marketers, not a 2027 one The EU AI Act's high-risk rules slipped to 2027. Article 50 disclosure still applies August 2, 2026, and it covers AI avatars, voiceovers, and chatbots.
- What actually governs your feed in 2026 The filter-bubble story took empirical hits, but feeds now shift specific attitudes, carry mostly AI slop, and face binding rules only in the EU.
- The agent stack, top to bottom Tools, skills, and agents: the three-layer mental model that explains every multi-agent system I've reviewed.
workshops
w-01
Skills Sprint
half-day · in person or virtualYour team leaves with three production-ready Claude skills they wrote themselves, plus the patterns to write the next ten. It's designed for engineering and ops teams.
request → w-02
Agent Architecture Review
full-day workshopBring your existing or proposed agent architecture. We work through it layer by layer using the Agent Stack model and list the failure modes and the governance hooks you're missing. Expect specific criticism of your design.
request → w-03
Responsible-AI Runtime Lab
2-day intensiveFor RAI and engineering teams in the same room. We turn your written policies into runtime checks (eval gates, trace filters, capability constraints), and you leave with running code.
request →For a workshop built around your team's situation, tell me the problem and I'll propose a session that addresses it.