https://caisi.dev/ https://caisi.dev/research/ https://caisi.dev/openclaw-2026/ https://caisi.dev/ai-tool-sprawl-q1-2026/ https://caisi.dev/ai-tool-sprawl-v2-2026/ https://caisi.dev/blog/ https://caisi.dev/blog/operating-notes/ https://caisi.dev/blog/openclaw/ https://caisi.dev/blog/sprawl-2026/ https://caisi.dev/blog/wrkr/ https://caisi.dev/blog/gait/ https://caisi.dev/blog/control-benchmarks/ https://caisi.dev/blog/governed-adoption/ https://caisi.dev/blog/ai-agent-governance/ https://caisi.dev/blog/glossary/ https://caisi.dev/authors/david-ahmann/ https://caisi.dev/blog/ai-engineering-control-problem/ https://caisi.dev/blog/repository-runtime-contract/ https://caisi.dev/blog/why-giant-instruction-files-fail/ https://caisi.dev/blog/from-skills-to-blueprints/ https://caisi.dev/blog/dark-factory/ https://caisi.dev/blog/isolated-warm-sandboxes/ https://caisi.dev/blog/parallel-agents-without-chaos/ https://caisi.dev/blog/visible-tests-are-not-enough/ https://caisi.dev/blog/proof-of-work-for-ai-generated-changes/ https://caisi.dev/blog/ai-engineering-maturity-model/ https://caisi.dev/blog/openclaw/stop-is-not-a-safety-control/ https://caisi.dev/blog/openclaw/discovery-is-necessary-not-sufficient/ https://caisi.dev/blog/openclaw/approval-and-proof-at-the-tool-boundary/ https://caisi.dev/blog/openclaw/what-openclaw-proves-and-what-it-doesnt/ https://caisi.dev/blog/sprawl-2026/evidence-before-exploitation/ https://caisi.dev/blog/sprawl-2026/approval-gap-is-a-proof-gap/ https://caisi.dev/blog/sprawl-2026/declared-agents-are-not-deployed-agents/ https://caisi.dev/blog/sprawl-2026/what-the-report-proves-and-what-leaders-should-do-next/ https://caisi.dev/blog/wrkr/most-dangerous-agent-is-uninventoried/ https://caisi.dev/blog/wrkr/mcp-config-is-part-of-security-boundary/ https://caisi.dev/blog/wrkr/ci-is-where-ai-tooling-becomes-infrastructure/ https://caisi.dev/blog/wrkr/npm-audit-for-ai-agents-is-the-wrong-endpoint/ https://caisi.dev/blog/gait/policy-after-the-tool-call-is-not-control/ https://caisi.dev/blog/gait/eslintrc-for-agent-behavior-is-an-incomplete-model/ https://caisi.dev/blog/gait/mcp-trust-needs-a-boundary-not-just-a-config/ https://caisi.dev/blog/gait/signed-traces-and-regressions-make-policy-real/ https://caisi.dev/blog/control-benchmarks/buyers-cannot-evaluate-agentic-control-clearly/ https://caisi.dev/blog/control-benchmarks/agent-action-risk-scenarios-minimum-test-set/ https://caisi.dev/blog/control-benchmarks/measure-control-efficacy-for-ai-agents/ https://caisi.dev/blog/control-benchmarks/proof-completeness-for-ai-agent-changes/ https://caisi.dev/blog/control-benchmarks/pilot-evaluation-framework-for-agentic-tools/ https://caisi.dev/blog/governed-adoption/what-platform-teams-must-standardize-before-ai-can-scale/ https://caisi.dev/blog/governed-adoption/right-response-to-shadow-ai-is-a-governed-path-not-a-ban/ https://caisi.dev/blog/governed-adoption/build-an-ai-operating-model-without-betting-on-one-stack/ https://caisi.dev/blog/governed-adoption/what-security-leaders-should-ask-before-approving-ai-in-delivery-workflows/ https://caisi.dev/blog/governed-adoption/sanctioned-ai-pathways-matter-more-than-blanket-restrictions/