What are AI agents and agentic systems, anyway?
A plain-English explanation of AI agents, agentic systems, and what makes them different from chatbots—plus the production lessons people skip.
I write about building and shipping AI applications from three main perspectives, also referred to as Tracks: engineering, business, and product.
Learn how posts are organised — and how to navigate them.
The 3 most important posts to start with, if you haven't already.
Showing 41 posts
Six approval gates, written as code instead of documentation. How PolicyForge makes sure a policy can't ship without you saying so.
A layered architecture sounds like diagram work. I built one so a broken component costs you a feature, never a lead.
How PolicyForge splits ingestion, ranking, generation, approval, and observability into five layers, and why that split keeps failures readable.
How PolicyForge catches AI compliance mistakes before they reach a submission, and routes them to the right human, every time.
Here are the exact latency, cost-per-unit, and uptime numbers PolicyForge holds itself to, and why loose budget language wasn't good enough.
Inside PolicyForge's approval workflow: why AI-generated policy recommendations never publish themselves, and how every decision gets recorded.
How PolicyForge generates multiple usable policy drafts instead of one, and keeps your data out of the process.
Inside the design decision that determines which compliance gap PolicyForge shows you first, and why.