Curious Cirkits Journal
Notes on AI.
Practical thinking on agentic architecture, institutional intelligence, and moving AI from demos into production.
Latest
Product thesis
Bring Your Own Agent: The Product Layer Above the Model
Agents can execute tasks. The harder product problem is preserving the user’s evidence, decisions, authority, and outcomes across the entire journey—even when the agent changes.
Product architecture
Building Curious Cirkits: What I’m learning about agentic products
What Jobs and Workout are teaching me about turning conversation into durable, measurable work—and building reusable agentic products.
Agentic AI is not a feature. It is an architecture.
The shift is not from software to chat. It is from workflows that move tasks to systems that can pursue outcomes—inside clear bounds.
Essay
From chatbots to delegated work
We are starting to delegate business processes—not just coding tasks. Are we ready to think differently about work and agency?
Product design
Build with intelligence. Operate with simplicity.
Invest intelligence upfront so simpler, faster, and less expensive models can operate the finished product reliably.
Agent design
A skill is a capability manual
A clean way to separate the agent, its tools, its state, its guardrails, and the human approval authority.
Product leadership
The real work of product building
Technology and architecture matter. But moving an idea forward also requires imagination, storytelling, and persistence.
Agent engineering
Measurement is the loop condition
Spec-driven development and agent teams only work when the goal has a verifiable exit condition.
Working with agents
How I work with Claude Code
Run several focused sessions like a small startup team, with Git history as the shared trace of how the work changes.