Blog
Guardrails, failure modes, pipeline design, and honest comparisons of the tools in this space.
September 29, 2026
GuardrailsThe interesting version of prompt injection isn't making a chatbot say something rude. It's making an agent with database credentials do something on an attacker's behalf.
Read more →September 22, 2026
GuardrailsApproval steps are the most common safety mechanism in agentic systems and the easiest to render meaningless. What separates a real gate from a speed bump.
Read more →September 17, 2026
ObservabilityThree ways to record what a system did, what each one costs, and why the choice is usually made by accident rather than on purpose.
Read more →September 09, 2026
ComparisonA neutral look at five mature orchestrators, what each one actually models, and the kinds of teams each design tends to suit.
Read more →August 31, 2026
GuardrailsThe useful question about an agentic workflow isn't what it's supposed to do. It's what it could do on its worst day, with the credentials you handed it.
Read more →