Practical writing on AI systems, knowledge infrastructure, and the mechanics of measurable delivery.
Latest dispatch
A task is not one thing. It's exploration, planning, and synthesis — each with different model requirements. Here's how to break the monolithic model assumption and route per-phase instead of per-task.
Archive
Operations, measurement, knowledge systems, and the systems behind technical work.
Default to the cheap model and justify escalation to the expensive one. Here's why flipping the burden of proof produces better AI systems — and better economics.
Read more
Most AI assistant architectures conflate two decisions: how hard should I think? and how much should I spend? Here's why they need to be separate control layers.
Read more
When departments optimize their own metrics, the system degrades. The mechanism isn't mathematical — it's psychological. Why metric ownership becomes identity, gaming is rational, and the fixes are …
Read more
When you build a custom AI assistant that knows your patterns and shapes how you think, you can lose track of what's real. Here's how to co-learn with …
Read more
How a 37-point Lighthouse improvement revealed the patterns that make human-AI partnerships work.
Read more
The JMP MCP Expert System started by answering questions. But the real request was never just an answer—it was executable workflows. This is the story of how session-to-JSL …
Read more
This is not just a chatbot story. It is a story about turning fragmented JMP support knowledge into a source-backed assistant surface with visible tool boundaries, provenance, and …
Read more
Fluent answers are not enough. If a user cannot inspect the evidence path behind an AI support answer, the product is optimizing for persuasion before trust.
Read more
I built a dashboard for my personal site. Then I made it public. This is what the new way of technical work looks like — AI-augmented, transparent, process-focused.
Read more
Obsidian taught a generation how to think in links. But the skill — linked thinking, atomic notes, emergent structure — has outlived the tool. The LLM is your …
Read more
Your KPIs do not measure exactly what you think they measure. Here is why interrogating your metrics—not just refining them—is the real work of DMAIC.
Read more
Framework fluency is easy to fake. The stronger hiring signal for AI systems roles is whether a candidate naturally reasons about constraints, failure modes, and causal chains.
Read more
How we publish practical AI systems writing with approval gates and factual rigor.
Read more