https://www.abisamsolutions.com/insights Field notes on AI reliability, evaluation, and data architecture from Abisam Solutions. en-us Wed, 02 Sep 2026 00:00:00 GMT https://www.abisamsolutions.com/insights/how-to-evaluate-an-ai-feature-before-you-ship-it https://www.abisamsolutions.com/insights/how-to-evaluate-an-ai-feature-before-you-ship-it Wed, 02 Sep 2026 00:00:00 GMT Evaluation A pass rate without a sample size is not a number. Here is the arithmetic that tells you how many eval cases you actually need, and why "we tested it manually" is a demo rather than a baseline. https://www.abisamsolutions.com/insights/what-to-log-to-debug-an-ai-system https://www.abisamsolutions.com/insights/what-to-log-to-debug-an-ai-system Wed, 02 Sep 2026 00:00:00 GMT Observability When a bad answer surfaces three weeks late, the answer alone tells you nothing. Write one trace record per model call, capturing the assembled context, and the failure becomes reproducible. https://www.abisamsolutions.com/insights/your-token-bill-is-a-design-problem https://www.abisamsolutions.com/insights/your-token-bill-is-a-design-problem Wed, 02 Sep 2026 00:00:00 GMT Cost Renegotiating your rate card moves an AI bill by percentages. Changing what you send on every call moves it by multiples. Here are the five design levers that actually control spend, with the arithmetic.