blog
Essays, posts & field notes
Longer essays and short observations on AI agents, governance, and what's working in production.
Looking for the newsletter? The Eval →
You can't copyright most of what your AI makes, and that's your smaller problem
Where AI and copyright actually stand for marketers as of mid-2026: why prompt-only output gets no protection after the Copyright Office's 2025 reports and the Thaler ruling, what the training-data cases (Bartz, Kadrey, Thomson Reuters, NYT) mean for the content you publish, and the fine print in your vendor's indemnification that decides who pays when an output infringes.
The EU AI Act is a 2026 problem for marketers, not a 2027 one
Everyone read the headline that the high-risk rules got delayed to 2027. The part that binds marketing teams — Article 50 transparency and disclosure — goes live August 2, 2026, and it covers your AI spokespeople, avatars, voiceovers, and chatbots. What's actually due, and what to do before the deadline.
What actually governs your feed in 2026
The filter-bubble story took real empirical hits, but recommendation feeds got harder to govern, not easier: feeds now demonstrably shift specific attitudes, AI slop is most of what's in them, and the only rules with teeth are the EU's. Where the evidence, the law, and the brand-safety problem actually stand.
Responsible AI Needs Architects, Not Just Advocates
The principles are written. The runtime isn't. Why the next wave of credible RAI leaders will ship middleware, not manifestos.
MCP and the new agent supply chain
Every MCP server you connect is third-party code your model trusts by reading. A pragmatic threat model and the controls that actually help.
Evals over vibes
If you can't say whether the new prompt is better than the old one, you're not iterating — you're vibing. The 30-minute fix.
Three orchestration patterns that aren't 'just call the LLM in a loop'
Routing, parallel fan-out, and orchestrator-workers — when to reach for which, with the tradeoffs that bite. Grounded in the subagents I actually run.
The agent stack, top to bottom
Tools, skills, agents — the three-layer mental model that explains every multi-agent system I've reviewed.
Protocols of Power: The Hidden Politics of Enterprise Multi-Agent Systems
My Cambridge dissertation on the protocols quietly becoming the governance standard for enterprise AI agents — Google's A2A and Anthropic's MCP — and why the choice to consolidate A2A under the Linux Foundation is a political act, not a technical one.
The Data Dilemma: Reconciling Privacy and Fairness in AI
Privacy and fairness are presented as opposing values in AI governance, but the choice between them is not actually a binary. A close look at GDPR exceptions, US self-regulation, the Amazon hiring algorithm, the Apple Card, and IBM's facial-recognition exit.
Beyond Principles: Examining the Conflicts in Implementing Responsible AI in Enterprise Structures
AI ethics has produced an oversupply of principles and a desperate shortage of mechanisms. A critical reading of Rakova et al. (2021) and Marchant (2021), with hard-won lessons from the FLA and IAB about why industry self-regulation usually fails.
Beyond the Hype: Navigating the Ethics of AI-Driven Marketing
Microtargeting and hypernudging raise valid autonomy concerns, but most of the public discourse is driven by simplistic narratives that overstate AI's persuasive power. A balanced framework — grounded in evidence — is what's actually missing.
ESG Echoes in AI Ethics: Drawing Lessons from Corporate Sustainability's Missteps for Effective Algorithmic Accountability
The ESG movement promised to align corporate behavior with social welfare and largely failed. AI ethics is on the same trajectory — unless we replace voluntary frameworks with enforceable structural reform.