--- title: Things I learnt building an enterprise AI Copilot description: A developer shares lessons learned building a six-tool enterprise AI Copilot. Key takeaways include: understanding requirements and data is harder than writing... canonical: https://daily.dev/posts/things-i-learnt-building-an-enterprise-ai-copilot-kgskqr95c twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Things I learnt building an enterprise AI Copilot | daily.dev og:description: A developer shares lessons learned building a six-tool enterprise AI Copilot. Key takeaways include: understanding requirements and data is harder than writing... og:url: https://daily.dev/posts/things-i-learnt-building-an-enterprise-ai-copilot-kgskqr95c og:image: https://api.daily.dev/og/posts/KgSKqr95C.png og:image:alt: Things I learnt building an enterprise AI Copilot og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Things I learnt building an enterprise AI Copilot **[Medium](https://daily.dev/sources/medium_js)** · 9 min read · 1 upvotes · 0 comments ## Summary A developer shares lessons learned building a six-tool enterprise AI Copilot. Key takeaways include: understanding requirements and data is harder than writing code, testing should be integrated throughout development (not just at the end), real-world production data rarely matches development datasets, good architecture is iterative, and the most effective enterprise AI systems often use a hybrid approach — combining deterministic logic with LLMs only when necessary. The hybrid architecture insight is particularly notable: raw data privacy concerns and cost/scalability issues make naive LLM-for-everything approaches problematic in enterprise settings, so a deterministic classifier with confidence scoring was used to invoke the LLM only for uncertain cases. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### When should I use a deterministic approach instead of an LLM for classifying data in an enterprise application? Use a deterministic approach when data privacy and cost are concerns, falling back to an LLM only when confidence is low. In one enterprise data governance tool, statistical features like null percentages, distinct values, and regex pattern matches produced weighted scores for classifying data quality dimensions, with a confidence threshold determining whether to accept the result or escalate to an LLM using only the computed statistics rather than raw data. _daily.dev surfaces engineering write-ups like this hybrid LLM design for teams weighing cost and privacy trade-offs._ ### Why does my data processing tool crash when given new production input files that worked fine during development? Production data often follows different formats or structures than the datasets used during development, causing tools built around specific assumptions to break. In one case, a working tool crashed because new input files followed a completely different, custom-built format than the test datasets, requiring the validation and preprocessing layers to be redesigned to support multiple input structures. _developers debugging production data mismatches can find similar real-world war stories on daily.dev._ ## Similar posts on daily.dev - [From LLM-First to Code-First: Lessons From Building Enterprise AI Systems](https://daily.dev/posts/from-llm-first-to-code-first-lessons-from-building-enterprise-ai-systems-ubac32c6o) · SD Times · 1 upvotes · 0 comments - [AI Skeptic to AI Pragmatist](https://daily.dev/posts/ai-skeptic-to-ai-pragmatist-ayf8wltkl) · Programming Digest · 28 upvotes · 0 comments - [Enterprise AI lessons learned from autonomous mobility](https://daily.dev/posts/enterprise-ai-lessons-learned-from-autonomous-mobility-1q8y439kq) · InfoWorld · 0 upvotes · 0 comments - [Reverse Engineering GitHub Copilot: What Every AI Product Manager Can Learn](https://daily.dev/posts/reverse-engineering-github-copilot-what-every-ai-product-manager-can-learn-l0nf4chfi) · Medium · 0 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#python](https://daily.dev/tags/python), [#llm](https://daily.dev/tags/llm), [#architecture](https://daily.dev/tags/architecture) [View this post on daily.dev](https://daily.dev/posts/things-i-learnt-building-an-enterprise-ai-copilot-kgskqr95c) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Things I learnt building an enterprise AI Copilot","url":"https://daily.dev/posts/things-i-learnt-building-an-enterprise-ai-copilot-kgskqr95c","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/things-i-learnt-building-an-enterprise-ai-copilot-kgskqr95c"},"datePublished":"2026-07-21T13:33:54.604Z","dateModified":"2026-09-14T08:41:00.989Z","description":"A developer shares lessons learned building a six-tool enterprise AI Copilot. Key takeaways include: understanding requirements and data is harder than writing...","image":"https://media.daily.dev/image/upload/s--1KxV4ohY--/f_auto/v1722860400/public/Placeholder%2007","thumbnailUrl":"https://media.daily.dev/image/upload/s--1KxV4ohY--/f_auto/v1722860400/public/Placeholder%2007","isAccessibleForFree":true,"articleSection":"Medium","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Medium","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/medium","url":"https://daily.dev/sources/medium_js"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/things-i-learnt-building-an-enterprise-ai-copilot-kgskqr95c","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai,python,llm,architecture","timeRequired":"PT9M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Medium","item":"https://daily.dev/sources/medium_js"},{"@type":"ListItem","position":3,"name":"Things I learnt building an enterprise AI Copilot"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/things-i-learnt-building-an-enterprise-ai-copilot-kgskqr95c#faq","mainEntity":[{"@type":"Question","name":"When should I use a deterministic approach instead of an LLM for classifying data in an enterprise application?","acceptedAnswer":{"@type":"Answer","text":"Use a deterministic approach when data privacy and cost are concerns, falling back to an LLM only when confidence is low. In one enterprise data governance tool, statistical features like null percentages, distinct values, and regex pattern matches produced weighted scores for classifying data quality dimensions, with a confidence threshold determining whether to accept the result or escalate to an LLM using only the computed statistics rather than raw data. daily.dev surfaces engineering write-ups like this hybrid LLM design for teams weighing cost and privacy trade-offs."}},{"@type":"Question","name":"Why does my data processing tool crash when given new production input files that worked fine during development?","acceptedAnswer":{"@type":"Answer","text":"Production data often follows different formats or structures than the datasets used during development, causing tools built around specific assumptions to break. In one case, a working tool crashed because new input files followed a completely different, custom-built format than the test datasets, requiring the validation and preprocessing layers to be redesigned to support multiple input structures. developers debugging production data mismatches can find similar real-world war stories on daily.dev."}}]} ```