Engineering notes
What we've learned securing AI agents, integrating enterprise systems and getting AI into production. Practical, and short.
Agent Safety Is an Engineering Problem
A policy document doesn't stop an agent mid-action. Identity, permissions, tool limits and logs do. Here is how to engineer them in.
Read →AI Agent Security: Runtime Controls That Actually Stop Things
Prompt filters catch some attacks. Runtime controls limit the damage when one gets through. A layered pattern for agents in production.
Read →Security · 6 min readSecure AI Agent Integration: Identity First
Before an agent touches your CRM or document library, decide who it is. A practical guide to agent identity with Okta or AWS IAM.
Read →Production · 7 min readAI Observability: What to Measure Before Go-Live
Uptime dashboards don't tell you an answer was wrong. Tracing, evaluation and cost signals that make an AI system operable.
Read →Integration · 5 min readEnterprise System Integration: Fix the Handoffs First
Most integration value sits in the handoffs between systems, not inside any one of them. How to find and fix the ones costing you most.
Read →Production · 6 min readMoving AI Agents From Prototype to Production
The demo worked. Production needs identity, evaluation, integration and an owner. A checklist for the gap in between.
Read →Have an integration problem?
30 minutes with an engineer, no slides. We agree the scope on the call and send pricing after.