Writing · Nº 041065 essays
The writing
Ideas from inside the work. Explore technology, entrepreneurship, and the practice of building things that last.
15 essays found · Page 1 of 2
Clear filtersAIHow to Tell If a Process Is Ready to AutomateNot every process should be automated. The signs a process is ready: it is documented, stable, high-volume, and rule-based. Automating a broken one just breaks faster.EngineeringHow to Trace an AI Workflow Run to Debug What Went WrongWhen an AI workflow produces a wrong result, the final output tells you nothing. Here is how per-step tracing lets you debug exactly which step went wrong and why.EngineeringHow to Test an AI Automation Workflow Before ProductionShipping an untested AI workflow straight to live data is how you break things at scale. Here is how to test automation with dry runs, fixtures, and a staging path.EngineeringHow to Handle Timeouts and Long-Running Workflow StepsA step that hangs can freeze an entire AI workflow. Here is how to set timeouts, handle long-running steps, and keep automation from stalling on a slow dependency.OperatingHow to Route Workflow Steps to the Right Model for CostUsing your best model for every step of an AI workflow is a waste. Here is how model routing sends each step to the cheapest model that can do the job well.OperatingHow to Cache Model Responses and Cut Automation CostsMost AI workflows pay full price to answer the same question twice. Here is how to cache model responses safely to cut token costs without serving stale answers.OperatingApproval Gates in Automation: When to Require a HumanNot every step in an AI workflow should run unattended. Here is how to place approval gates so high-stakes actions pause for a human without killing throughput.EngineeringHow to Version and Roll Back an Automation WorkflowWhen an AI workflow change breaks in production, you need to roll it back in seconds. Here is how to version automation workflows so a bad deploy is never permanent.EngineeringWebhook vs Polling for Triggering Automation WorkflowsShould your AI workflow be triggered by a webhook or a polling loop? Here is how to decide, when each one wins, and the failure modes both hide from you.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.OperatingHow to Set Token Budgets Per Workflow RunA runaway AI workflow can burn a fortune in tokens before anyone notices. Here is how to set per-run token budgets that cap spend and stop the bleed automatically.