HARRISON NDEKE / TECHNICAL JOURNAL
Practical engineering notes on AI systems, automation, and the boring-but-important work that keeps a workflow dependable after launch.
FEATURED NOTE / 001API design / batch processing / idempotency / workflow reliability
For a partial-success batch API, create a durable ledger before the call, classify every item from the response, and retry only items the provider explicitly left unprocessed or identifies as transiently failed. Treat a timeout or missing response as unknown, not failed: reconcile by an idempotency key or a read-back query before sending anything again.
Read the field noteTHE ARCHIVE
Each note starts with a real problem, separates confirmed practice from opinion, and includes the trade-offs a production team needs to consider.
01API design / batch processing / idempotency / workflow reliability
A practical pattern for item-level outcome ledgers, selective retries, reconciliation, and idempotency when a batch API completes only part of a request.
Read note
02AI agents / agent evaluation / tool calls / external effects
A practical guide to testing an AI agent's full trajectory, approvals, tool calls, external effects, and final system state before release.
Read note
03AI agents / tool calls / JSON Schema / authorization
A practical guide to validating AI tool arguments, enforcing business rules and authorization, and testing consequential agent actions before execution.
Read note
04backpressure / n8n / queue mode / workflow scaling
A practical guide to concurrency limits, bounded queues, retry budgets, load shedding, and overload monitoring for n8n workflows and AI agents.
Read note
05prompt injection / RAG security / AI agents / tool security
A practical architecture for treating retrieved content as untrusted data, constraining agent tools, validating actions, and testing poisoned sources.
Read note
06observability / n8n / workflow monitoring / automation reliability
A practical observability guide for connecting automation logs, metrics, traces, alerts, and synthetic checks to verifiable business outcomes.
Read note
07idempotency / n8n / workflow reliability / duplicate prevention
A practical idempotency pattern for preventing duplicate messages, records, charges, and other side effects when automation events are retried.
Read note
08AI agents / human approval / tool security / risk controls
A practical risk model for deciding which AI agent tool actions can run automatically, which need human approval, and what an approval gate must verify.
Read note
09n8n / webhooks / security / API security
A practical security checklist for n8n webhooks covering authentication, request validation, replay protection, resource limits, safe responses, and production testing.
Read note
010n8n / reliability / automation / error handling
Learn when to retry an n8n failure, when to use a fallback, how to configure an error workflow, and when to send work to human review.
Read noteEDITORIAL STANDARD
There are no generic AI roundups here. Every Field Note is written to help a builder make a clearer technical decision: how to model failure, when to introduce review, what should be measured, and where a clean demo is not enough.
Discuss a system