In this episode of Root Cause we sit down with Evan Marshall, co-founder and CTO of Ito.ai, to get to the root cause of where the honest answer sits on the AI adoption spectrum, somewhere between the teams that won't let AI near their codebase and the ones that let it ship with barely a human in the loop. Most of us are somewhere in the middle, moving around and quietly wondering if we've got it right. This conversation is meant to help you gauge where you actually are, and where you want to be. Evan studied computer science at MIT and spent four years building privacy infrastructure for zero-knowledge blockchains, where a slipped bug is a disaster. That experience pushed him toward quality: he now builds a QA agent that actually runs your application and checks the code before it merges.
We dig into the pressure on Bay Area companies to claim a "software factory" while their engineers might still be shipping like everyone else, why free code generation can at most double you if coding was half your job (and it never was), and the plastic age of coding, where code is generated to be thrown away. Evan shares what a sane monthly AI spend actually looks like and his advice for integrating AI into your software development lifecycle: treat the SDLC as an optimization problem, because once generation is free, the bottleneck moves to verification and quality control. And when someone says "I can just do that with Claude Code", the work that follows is environments that have to heal themselves, caching that breaks on the next migration, and memory layers. As Evan puts it: we spent the tokens so you don't have to.
00:00 Pre-show Banter and the Cats 00:59 Introduction of the Episode and the Guest
03:00 Is the Bay Area Really an AI Echo Chamber?
05:45 The Pressure to Claim a Software Factory
07:36 How Much Does AI Actually Speed You Up?
10:23 How Ito Uses AI in Its Own Engineering
14:13 The Plastic Age of Coding
15:53 Can the Software Factory Be Productized?
18:39 Subscriptions vs API: The Real Cost of Tokens
20:19 Open Models and Switching Between Them
21:25 Are the Models Plateauing?
22:59 "I Can Just Use Claude Code": Where the Hard Parts Hide
26:19 The Bottlenecks You Don't See: Environments, Caching, Memory
28:55 Will QA Engineers Be Replaced?
31:39 Replicators, Waymos, and 3D Printers: Why Change Takes Time
36:49 Why SaaS Isn't Dead and How Much You Should Spend
39:23 Advice for Integrating AI into Your SDLC
42:50 Question from Barry O'Reilly: Building a Computer in 1822
44:31 Shared Reality, Social Media, and Cooperation
50:25 Book Recommendations
53:54 Question for the Next Guest and Closing
Find Evan at:
• LinkedId: [https://www.linkedin.com/in/evan--marshall/](https://www.linkedin.com/in/evan--marshall/)
• his startup - [https://www.ito.ai/](https://www.ito.ai/)
Find me (Nune) at:
LinkedIn:
[https://www.linkedin.com/in/nisabek/](https://www.linkedin.com/in/nisabek/)Substack:
[https://www.thoughtfultechnologist.com/](https://www.thoughtfultechnologist.com/)