Blog
Architecture decisions, build transparency, and lessons from building governed AI systems with an AI workforce. We write three types of content: architectural thinking (the ideas), product transparency (the build), and operational insight (the lessons).
Architectural Thinking
Long-form arguments about how AI governance, knowledge systems, and organizational cognition should work. Grounded in research. Not product pitches.
Coming Q2 2026
AutoGen, CrewAI, LangGraph — every major framework ships ungoverned agents. We assess six agent memory frameworks (Letta, Mem0, Zep, Hindsight, LangMem, Cognee) against six governance capabilities. None pass. Here is the structural gap, why prompt-level guardrails do not close it, and what architectural enforcement actually looks like — starting with Cedar ABAC at the query boundary.
Coming Q2 2026
Every agent memory system stores knowledge at one level. LEAPCortex stores every knowledge object at four resolution levels — one-line status, structured summary, detail, and raw source — and serves the right depth based on context window capacity, task requirements, and Cedar-enforced authorization. A technical deep-dive into the ResolutionVariants architecture and the Context Assembly Service that makes dynamic resolution selection possible.
Product Transparency
What we built, why we built it, and what we learned — including what did not work.
Coming Q2 2026
How one founder and a governed AI workforce produced 169 specification documents, 17 patent applications, 75 wireframes, and 28 governance standards. The operating model, the failures (seven governance violations in the first two weeks), the structural fixes (five Operational Directives), and what we learned about quality at velocity.
Coming Q2 2026
Following the LEAPCortex build from specification to production. Architecture decisions, performance benchmarks, integration challenges, and the real cost of building governed AI infrastructure from a 169-document specification suite.
Operational Insight
Practical lessons from operating LEAPWare with a governed AI workforce.
Coming Q2 2026
A technical deep-dive into using Amazon Cedar as the ABAC policy engine for AI agent governance. How we evaluate every agent action against declarative policies before execution with sub-10ms latency. Why Cedar over OPA, Casbin, or custom policy engines. The trade-offs we found and the patterns that worked.