Writing
I write about product craft, designing AI features, and keeping agents safe.
Selected
2026
Architectural patterns for shipping non-deterministic LLM features safely: confidence-gated rendering, out-of-band degradation, and structured telemetry.
Related work: AI-assisted activity creation
2026
The four constraints required for secure agent authorization, demonstrated in a 2,000-line reference implementation.
Related work: A2A↔MCP bridge
2023
A structured framework for connecting organizational mission, strategy, and execution. Two parts, with a worked example.
Related work: Uptempo Campaign Management
All writing
2026
How to ship reliable AI form suggestions by gating the interface on confidence scores, keeping fallback paths unassisted, and logging accept/reject telemetry from day one.
2026
Why boolean approval prompts fail when AI agents call destructive tools, and how parameter binding, bounded intent, consent atomicity, and cryptographic delegation solve the authorization problem in practice.
2026
The single-use property has two distinct layers. Most published designs enforce only the credential layer, leaving human approval vulnerable to mint replay. Here is how to close the gap across distributed replicas.
2026
A code-anchored walkthrough of how ActionAuth realizes three-layer delegation, durable replay state across distributed replicas, and the two carriers that move human approvals across single-agent and multi-agent systems.
2023
Applying the Mission, Strategy, and Work framework to a hypothetical multi-year logistics strategy for Zipline.
2023
Most organizations operate with high-level board metrics and a disconnected backlog of projects. A practical framework for connecting mission, strategy, and execution.