Posts are chapters. These are the programs they belong to. I ship the system, the rule, or the talk first. The post is what got written down after.
The through-line lives on Projects: memory, tools, deterministic constraints, evaluation, observability, security, and human control.
1 · Memory for AI agents
Memori, architectures, retrieval, decay, identity, evaluation.
- Forging AI’s lasting memory
- Memori architecture
- Memori recursive design
- Talk: Memory is the Agent
- Artifact: Memori
- Artifact: Chronicle
2 · Reliable agent systems
Luffy, tool design, control planes, observability, failure modes.
- Designing CLI tools for AI agents
- Hard constraints belong in code
- Building Luffy
- Artifact: Luffy
- Artifact: Idea Generator
3 · AI × security
Agent attacks, fraud, identity, social engineering, behavioral security.
These PRs are the same stack. Identity, false positives, and constraints, written as detection rules the way Memori and Luffy write them for agents.
- The Browser Is the New Operating System examines the browser as the capability broker for identity, data, devices, and isolation.
- Potential Proxy Execution Via SetupUGC (rule).
setupugc.exeas a signed LOLBin on Windows 10/11 and Server 2025. - Exclude Kea DHCP from CAP_NET_RAW. The building block was firing on a legitimate
/usr/sbin/kea-dhcp4. - AWS discard_regex values with spaces. Unquoted Security Hub values split as extra arguments.
4 · AI-native software engineering
Coding agents, architecture, productivity, staff+ engineering, evaluation.