Running Subagents in Parallel Without One Failure Sinking the Whole Run
A fan-out / fan-in design for running several subagents in parallel, covering token budgeting, a result contract, and partial-failure handling. Includes an implementation where one branch can fail without stopping the rest, plus measured numbers.
Putting Cloudflare AI Gateway in Front of Claude Made the Numbers I Needed Disappear — Field Notes on Instrumentation
After putting Cloudflare AI Gateway in front of Claude API, here is where I actually got stung — cost attribution, semantic-cache false hits, fallback quietly lowering quality, and budgets that don't really stop anything — with the code I used to fix each.
Grounding Claude on Your Own Knowledge Base with search_result Blocks
How to stop your own RAG setup from losing track of which article it cited, using Claude's search_result content block and structured citations — with real numbers from running it across four sites.
When Claude API Extracts the Wrong Value With Full Confidence — Designing the Verification Layer
When you extract invoices or contracts with Claude API, the scariest failure isn't an exception — it's plausible-but-wrong JSON. Here is how I build a verification layer that catches silent extraction errors with schema checks, arithmetic reconciliation, and dual-extraction agreement, in TypeScript.
Making the Numbers Add Up in a Multi-Tenant Claude API SaaS — Field Notes on Isolation and Cost Attribution
The first thing that breaks when you make a Claude API SaaS multi-tenant is the month-end reconciliation. Here are field notes on a single metering chokepoint, atomic counters, reconciling against Anthropic's bill, and proving tenant isolation with adversarial tests — with production TypeScript.
Keep a Decision Rationale Ledger for Autonomous Agents — So You Can Explain 'Why' Later
When an autonomous agent takes hard-to-reverse actions like a production deploy or a bulk delete, capture the chosen option, rejected alternatives, and assumptions in a structured ledger. Includes structured output, an append-only log, and tiering by impact.
Taming Token Bloat in Long-Running Agents with Context Editing and the Memory Tool
For long-running agents whose input tokens balloon as tool results pile up, here is how to pair context editing with the memory tool and measure the savings with count_tokens, including a working backend implementation.
Trusting Claude's Structured Output in Production — Validation Gates and Repair Loops
When Claude's structured output breaks 'occasionally' in production, combine tool-use enforcement, a schema validation gate, a single repair loop, and a graceful degradation fallback to eliminate broken JSON from your operations — with working TypeScript code.
Confirm Your Model Actually Responds Before a Scheduled Run Begins
A model you configured can be gone before your nightly job even wakes up. Tell retirement, withdrawal, and regional restriction apart with a single startup probe, then rewrite the run config to an eligible model — with complete, working TypeScript.
Centralizing the anthropic-beta Header So a Retired Beta Won't Kill Your Batch
Scattered anthropic-beta headers turn a beta retirement or GA graduation into a 400 that takes down an entire batch. A small capability registry, a startup preflight, and tiered fallback keep your pipeline running across feature generations.
Letting Claude Read Live Pages: Implementing web_fetch Without the Footguns
How to pull the actual text of official pages and PDFs straight into Claude's context with the web_fetch tool. Covers the URL-validation rule that trips everyone up, the settings that keep token costs down, and why fetch errors arrive as HTTP 200 — based on what I hit running it in production.
Record Which Model Actually Answered — Attestation Logging for Headless Pipelines
Persist the model field and usage from every API response so you can detect when the served model differs from the one you requested, and reconcile per-model cost ahead of the usage credits change.