Writing · Nº 041065 essays
The writing
Ideas from inside the work. Explore technology, entrepreneurship, and the practice of building things that last.
45 essays found · Page 3 of 4
Clear filtersThesisAI-Native Support Resolves, a Bolted-On Chatbot DeflectsAn AI-native support product resolves the issue. A chatbot bolted onto a legacy helpdesk just deflects tickets. Here is the difference buyers keep missing.OperatingQuestions to Ask Before Buying an AI-Native ProductThe questions to ask an AI-native vendor before you buy. Cut through the marketing and find out if the model is load-bearing or bolted on for the pitch.EngineeringLLM as Judge: When to Trust It for AI EvaluationUsing an LLM as judge to score AI output is fast and scalable, but it is biased and gameable. When to trust LLM-as-judge for evaluation and when to bring in a human.ThesisAI-Native Underwriting vs a Rules Engine With AI AddedAI-native underwriting reasons over the whole file. A rules engine just gates it. Here is why native underwriting beats bolting a model onto legacy scorecards.EngineeringWhy AI-Native Products Need a Different Data ModelAI-native products need a different data model than normal software. Why the schema has to store provenance, versions, and model reasoning, not just final values.EngineeringOffline Evals vs Production Monitoring for AIOffline evals and production monitoring measure different things for an AI feature. You need both. Here is what each catches, what each misses, and how they work together.ThesisThe Myth That AI-Native Means Firing the ProfessionalsAI-native in regulated fields does not mean replacing the doctor, lawyer, or accountant. It means changing what they do. Here is the myth and the reality.ThesisWhy Most AI Wrappers Fail Within a YearWhy do AI wrappers fail? Thin wrappers around a model get out-executed and commoditized fast. What separates a durable AI product from a disposable wrapper.EngineeringHow to Build a Golden Eval Set for an AI FeatureA golden eval set is the fixed, labeled test cases that tell you if an AI feature got better or worse. Here is how to build one before you tune a single prompt.ThesisThe Healthcare AI Scribe Retrofit TrapAn AI scribe stapled to a 2010 EHR still leaves the clinician inside the form. Here is why the healthcare AI retrofit caps out and what native design fixes.EngineeringWhat Actually Changes in AI-Native ArchitectureAI-native architecture is not a chatbot on a normal app. Here is what actually changes in the data model, control flow, and error handling when the model is core.EngineeringAI-Native Dev Environment vs Autocomplete Bolted OnAn AI-native dev environment plans and edits across the whole repo. Autocomplete just predicts the next token. Here is why that gap decides real throughput.