Writing · Nº 041065 essays
The writing
Ideas from inside the work. Explore technology, entrepreneurship, and the practice of building things that last.
36 essays found · Page 3 of 3
Clear filtersAICommon Mistakes When Buying AI AgentsThe most common mistakes buying AI agents: falling for the demo, ignoring failure modes, over-scoping autonomy, and skipping the audit trail. How to avoid each.AIWhat Makes an AI Agent ReliableAI agent reliability comes from architecture, not a smarter model. Here are the five design choices that separate a dependable agent from a confident guesser.EngineeringWeb Scraping Mistakes That Get You BlockedThe common web scraping mistakes that get you blocked: bad rate limits, obvious fingerprints, ignoring soft blocks, and treating every target the same way.EngineeringHow to Handle Rate Limits and Backoff in AI IntegrationsThird-party APIs will throttle your AI workflows. Here is how to handle rate limits with backoff, jitter, and queuing so integrations stay reliable under load.EngineeringWhy Your AI Automation Needs a Dead Letter QueueWhen an AI task fails every retry, it should not vanish. A dead letter queue catches the failures your workflow could not handle so nothing is ever silently lost.AIHow to Evaluate an AI Agent Before You Deploy ItA practical checklist to evaluate an AI agent before deployment: reliability, guardrails, audit, tool scope, and fallback behavior, not just the demo.EngineeringSet RPO and RTO Targets Before You Design DRRPO and RTO decide your entire disaster recovery design. Set both numbers before you buy anything, so you build the recovery your self-hosted apps actually need.EngineeringPin the Model Version in Production, Don't Float ItLetting your AI product float to the latest model version means your behavior changes without warning. Why to pin the model version in production and upgrade on purpose.AIWhy AI Agents Fail in Production (and How to Fix It)AI agents fail in production for a handful of predictable reasons. Here are the real failure modes and how to design agent reliability from the start.EngineeringHow to Handle Partial Failure in a Multi-Step WorkflowA multi-step workflow that fails halfway leaves your systems inconsistent. Here is how to handle partial failure with compensation, checkpoints, and clean rollback.EngineeringIdempotency in AI Workflows: Retry Without Double-ChargingIdempotency is what lets an AI workflow retry a failed step without sending two emails or charging a card twice. Here is how to build it in from the start.EngineeringTest Your Backups With Restore Drills, Not FaithA backup you have never restored is a guess. Here is how to run restore drills on a schedule so your self-hosted recovery actually works when it counts.