Перейти к содержимому

Episode 3: How Okta Uses Cost Models to Make Better Engineering Decisions

Frugal AI

0:00 / 0:00

Episode 3: How Okta Uses Cost Models to Make Better Engineering Decisions

73 просмотра · 5 дн. назад
Frugal AI
52 подписчика
73 просмотра · 5 дн. назад
What should a new feature cost before it ships? Mariano Szklanny of Okta builds the cost models that answer that question, and in one case his model caught a planned architecture that would have doubled fixed costs before a line of it went live. Mariano has worked at the intersection of engineering and economics since his early days at Auth0, when AWS costs spiked and nobody knew why. In this episode of Shift Left FinOps, he and host Mike Weider (CEO, Frugal) walk through how Okta models cost early, decides which features need a model at all, and checks the assumptions once real usage arrives. In this episode: How a cost model exposed double fixed costs in a new feature, and the architecture change that let it ship with healthy margins Model unit cost first, then layer on low, medium, and high adoption scenarios Why the AWS pricing calculator is only a first step for pricing decisions Swapping spreadsheets for interactive HTML models that engineers can challenge and correct How AI cut modeling time from one to two months down to about two weeks Two questions that triage which features need a model: new infrastructure, and more traffic Telling the data team which usage metrics to capture before launch Cost as a business enabler: giving finance the confidence to ship AI agent features fast Advice for teams starting out: get cost and usage data somewhere you can query, and build the model for the person who has to use it Chapters: 00:00 Intro 00:32 The cost model that changed an architecture 02:40 Mariano's role and what Okta does 04:25 From engineering ops to FinOps 05:58 What a cost model is 07:59 What a useful model should answer 09:07 Cost models vs budgets, forecasts, and pricing calculators 11:19 Building a model: unit cost, then usage scenarios 14:01 How AI changed the process 16:27 From two months to two weeks 17:31 Deciding which features get a cost model 20:57 Who is involved and when 23:26 Instrumenting usage and monitoring after launch 27:34 Balancing cost against security, reliability, and growth 30:34 Advice for teams starting out 34:11 Wrap up Guest: Mariano Szklanny, Okta https://www.okta.com 👉 Book a founder-led demo to see it work for you: https://frugal.co/book-a-demo 🎁 Turn inefficient code into cloud savings: https://frugal.co/ 🤝 Connect with Us:   / frugalco   https://frugal.co/ #FinOps #ShiftLeft #CloudCosts #CostModeling #AWS