https://catylai.com/blog/ Field notes on AI-assisted engineering and cloud architecture. en Thu, 24 Sep 2026 00:00:00 +0000 https://catylai.com/blog/one-door-into-the-warehouse/ https://catylai.com/blog/one-door-into-the-warehouse/ Steward gives Claude a single read-only tool over the curated warehouse, with field-level authorization compiled to Cedar. Here is why I built it that way instead of wiring up one connector per product. Thu, 24 Sep 2026 00:00:00 +0000 agents architecture patterns https://catylai.com/blog/an-adr-is-only-as-good-as-what-reads-it/ https://catylai.com/blog/an-adr-is-only-as-good-as-what-reads-it/ A decision record that only lives as prose is something an agent has to be told to go open. The ones that survive are the ones also wired into a table, a script, or a test the agent's normal workflow already touches. Tue, 22 Sep 2026 00:00:00 +0000 architecture patterns claude-code https://catylai.com/blog/bedrock-access-three-gates/ https://catylai.com/blog/bedrock-access-three-gates/ A Bedrock AccessDeniedException can mean three different things, so we built a nightly discovery job and a runbook that checks each gate in order. Tue, 22 Sep 2026 00:00:00 +0000 aws architecture devex https://catylai.com/blog/credential-nobody-watched/ https://catylai.com/blog/credential-nobody-watched/ A shared service-account token expired silently and took a production integration down for 40 hours. The fix that mattered wasn't rotation. It was making four causes of the same error distinguishable. Tue, 22 Sep 2026 00:00:00 +0000 aws patterns devex https://catylai.com/blog/deleting-five-review-agents/ https://catylai.com/blog/deleting-five-review-agents/ A five-specialist review fan-out looked thorough. One session cost $5.88 and 905 seconds for zero findings, and that's what actually changed the design. Tue, 22 Sep 2026 00:00:00 +0000 claude-code agents ci https://catylai.com/blog/designing-for-the-non-programmer-maintainer/ https://catylai.com/blog/designing-for-the-non-programmer-maintainer/ A change-tier system and a verifier that fails risky edits even when every test passes, because a green suite is not evidence the change was safe. Tue, 22 Sep 2026 00:00:00 +0000 architecture devex ai https://catylai.com/blog/deterministic-math-as-a-tool-call/ https://catylai.com/blog/deterministic-math-as-a-tool-call/ Why I moved every numeric answer out of model prose and into a calculator tool call, and what a data-quality investigation taught me about trusting the output. Tue, 22 Sep 2026 00:00:00 +0000 architecture patterns agents https://catylai.com/blog/field-level-authorization-cedar/ https://catylai.com/blog/field-level-authorization-cedar/ Row-level access control tells an agent nothing exists. Section-level tells it the entity exists with fewer fields, and that difference changes what an agent can safely ask. Tue, 22 Sep 2026 00:00:00 +0000 claude-code agents architecture https://catylai.com/blog/forcing-the-model-to-show-its-math/ https://catylai.com/blog/forcing-the-model-to-show-its-math/ An LLM cannot be trusted to self-report whether it actually did math, so a Stop hook checks mechanically whether a real calculation tool ran first. Tue, 22 Sep 2026 00:00:00 +0000 claude-code agents patterns https://catylai.com/blog/forty-seven-adrs-in-six-weeks/ https://catylai.com/blog/forty-seven-adrs-in-six-weeks/ A decision log is valuable because it preserves disagreement with your past self, illustrated by ADRs that reversed their own measured evidence in place. Tue, 22 Sep 2026 00:00:00 +0000 architecture patterns devex https://catylai.com/blog/green-pipeline-not-evidence/ https://catylai.com/blog/green-pipeline-not-evidence/ Two load-bearing CI gates ran green for months without comparing anything, because the base image had no git and nobody noticed the difference between passing and not running. Tue, 22 Sep 2026 00:00:00 +0000 patterns devex build https://catylai.com/blog/infrastructure-new-client/ https://catylai.com/blog/infrastructure-new-client/ The architecture diagram for our shared AWS platform names two human actors and lists an AI coding agent's own cloud egress as a third. That's a trust-model decision, not a diagram detail. Tue, 22 Sep 2026 00:00:00 +0000 aws architecture claude-code https://catylai.com/blog/judge-that-never-fires/ https://catylai.com/blog/judge-that-never-fires/ A quality check the model can choose to invoke isn't weak enforcement. It's no enforcement, and the corpus behind it has a measured number to prove it. Tue, 22 Sep 2026 00:00:00 +0000 claude-code agents patterns https://catylai.com/blog/migrating-off-a-hand-rolled-agent-loop/ https://catylai.com/blog/migrating-off-a-hand-rolled-agent-loop/ When a vendor SDK actively maintains the parts of an agent loop you built by hand, the build-vs-adopt math changes even if your version already works. Tue, 22 Sep 2026 00:00:00 +0000 architecture agents claude-code https://catylai.com/blog/one-gateway-nine-backends/ https://catylai.com/blog/one-gateway-nine-backends/ Why an internal MCP aggregation gateway hand-rolls raw HTTP instead of the standard SDK client, and how it disambiguates tool names across eight backends. Tue, 22 Sep 2026 00:00:00 +0000 architecture patterns build https://catylai.com/blog/shared-server-utils-two-languages/ https://catylai.com/blog/shared-server-utils-two-languages/ Running the same MCP service shape in TypeScript and Rust only works if you decide, in writing, which parts are a contract and let everything else diverge. Tue, 22 Sep 2026 00:00:00 +0000 architecture patterns agents https://catylai.com/blog/the-prompt-cache-floor-the-cheap-tier-never-cleared/ https://catylai.com/blog/the-prompt-cache-floor-the-cheap-tier-never-cleared/ A two-tier triage pipeline assumed its cheap model's shared prompt was cached. Real token telemetry showed it never once was. Tue, 22 Sep 2026 00:00:00 +0000 architecture ai patterns https://catylai.com/blog/what-a-bot-vote-is-allowed-to-decide/ https://catylai.com/blog/what-a-bot-vote-is-allowed-to-decide/ The architecture decision trail behind letting an AI code-review job cast a real merge approval, and the explicit line we drew around what that vote can never do alone. Tue, 22 Sep 2026 00:00:00 +0000 architecture ai build https://catylai.com/blog/delegating-to-agents-without-losing-the-thread/ https://catylai.com/blog/delegating-to-agents-without-losing-the-thread/ Fanning work out to parallel agents is easy. Staying responsible for the result is the part that needs a method. Mon, 21 Sep 2026 00:00:00 +0000 claude-code agents https://catylai.com/blog/hosting-a-claude-code-marketplace/ https://catylai.com/blog/hosting-a-claude-code-marketplace/ It works, but three undocumented-feeling constraints will bite you. Here they are before they do. Mon, 21 Sep 2026 00:00:00 +0000 claude-code plugins https://catylai.com/blog/static-exports-do-not-forgive/ https://catylai.com/blog/static-exports-do-not-forgive/ Four ways a Next.js static export broke during a rebuild, and what each one was actually telling me. Mon, 21 Sep 2026 00:00:00 +0000 nextjs build https://catylai.com/blog/what-outlives-the-terraform/ https://catylai.com/blog/what-outlives-the-terraform/ Moving off AWS is mostly deletion. The risk is entirely in the handful of things deletion does not cover. Mon, 21 Sep 2026 00:00:00 +0000 infrastructure dns