CrownCodes ยท WritingField notes on agent architecture, evaluations and systems that fail safely.https://www.crowncodes.com/The hard part of agentic AI is deciding what an agent can dohttps://www.crowncodes.com/writing/deciding-what-an-agent-can-do/https://www.crowncodes.com/writing/deciding-what-an-agent-can-do/Agent authority belongs in software, not in a prompt. Five questions to answer before you hand a workflow to a model, each with an artefact you can build.Tue, 29 Sep 2026 00:00:00 GMTEvaluating an invoice-matching agent before you trust ithttps://www.crowncodes.com/writing/evaluating-an-invoice-matching-agent/https://www.crowncodes.com/writing/evaluating-an-invoice-matching-agent/How I would design an evaluation for an agent that matches payments to invoices: cost-weighted outcomes, stratified cases, deterministic graders first, and CI gates.Tue, 29 Sep 2026 00:00:00 GMTSonnet 5.5 ends forced tool use. Here's how to keep structured output reliablehttps://www.crowncodes.com/writing/forced-tool-use-after-sonnet-5-5/https://www.crowncodes.com/writing/forced-tool-use-after-sonnet-5-5/Sonnet 5.5 returns a 400 for tool_choice any or tool. Moving to auto plus strict tools changes how extraction fails, so the pipeline needs a missing-call check, a bounded retry, a fallback and an eval.Tue, 29 Sep 2026 00:00:00 GMTThe model gateway: routing, timeouts, retries and what they costhttps://www.crowncodes.com/writing/model-gateway-routing-timeouts-retries/https://www.crowncodes.com/writing/model-gateway-routing-timeouts-retries/Product code should make one call with one contract. A thin gateway underneath owns routing, timeouts, retries, fallbacks and the log line that tells you what each call cost.Tue, 29 Sep 2026 00:00:00 GMTA permission matrix for an invoice-matching agenthttps://www.crowncodes.com/writing/permission-matrix-for-an-invoice-agent/https://www.crowncodes.com/writing/permission-matrix-for-an-invoice-agent/How to decide, write down and enforce in code what a payment-matching agent may do, from reading invoices to never touching vendor bank details.Tue, 29 Sep 2026 00:00:00 GMTReconciliation is the original agent evalhttps://www.crowncodes.com/writing/reconciliation-is-the-original-agent-eval/https://www.crowncodes.com/writing/reconciliation-is-the-original-agent-eval/Finance ops worked out how to trust automated processes long before LLM agents. A map from reconciliation controls to agent-harness equivalents, and where teams get each one wrong.Tue, 29 Sep 2026 00:00:00 GMTThe prompt said no internet: what three months of agent escapes teach buildershttps://www.crowncodes.com/writing/the-prompt-said-no-internet/https://www.crowncodes.com/writing/the-prompt-said-no-internet/Between July and September 2026, agents at several labs crossed boundaries that existed only as text. Each failure maps to a software control, and the vendors are moving the same way.Tue, 29 Sep 2026 00:00:00 GMTThreat modelling a tool-using agent in one afternoon, before it shipshttps://www.crowncodes.com/writing/threat-modelling-an-agent-before-deployment/https://www.crowncodes.com/writing/threat-modelling-an-agent-before-deployment/A practical threat model for an LLM agent with tools and MCP servers: draw the boundaries, walk the threats, put each control in code, and red-team it before launch.Tue, 29 Sep 2026 00:00:00 GMT