Article
The Hugging Face intrusion: why catching a rogue agent on day three isn’t enough.
17,600 actions before anyone caught it. Detection worked, but detection isn’t prevention. Here’s the boundary that has to come first.
Article
Why we build agents that go to the data, not the other way around.
Most AI deployments fail because teams spend 80% of their time moving data. We don’t do that, and it changes everything about how fast you can ship.
Paper
The hybrid pipeline pattern: deterministic agents vs. LLM calls.
1 API call per run instead of 30. How deterministic agents and a single LLM supervisor reach production scale.
Article
Governance built in: how D8:SCOUT enforces policy without slowing the pipeline.
Security at the agent level behaves differently than security bolted on at the perimeter.
Paper
Data lineage at scale: full traceability without a data catalog rebuild.
D8:CAT maintains origin-to-output lineage on every record, as a byproduct of normal pipeline operation.
Article
Human-in-the-loop by design: where agents stop and people decide.
Agentic AI doesn’t mean unattended AI. Here’s how we wire checkpoints into every pipeline from day one.
Paper
From 5% sampling to 100% coverage: loan audit automation at Kitsap Credit Union.
A production deployment case study. 100% portfolio coverage. Every loan. Every night. No migration.
Video
Pipeline walkthrough: how a funded loan moves through 6 agents in under 3 seconds.
Watch a single loan move through all 6 agents: real timestamps, real data, real exception handling.