What This Blog Is About
Notes from putting AI agents into a real Fortune-500 iOS review loop and measuring it, plus the architecture calls behind a decade of shipped apps.
I put Claude Code subagents, MCP servers, and browser automation into the actual code-review and test-generation loop for a multinational iOS team, and measured what it did: about 30 minutes saved per engineer per day, review-cycle time down, review quality held. Most of what gets written about AI in engineering is written by people who have not had to defend the number afterwards. That is the gap this blog exists to fill.
What you’ll find here
AI agents in engineering teams that actually ship. Where an agent is trusted and where a human stays in the loop, how to encode review taste into one instead of letting it optimize a proxy like “tests pass”, and what breaks when you skip that. The hard part is never getting the agent to produce output.
iOS architecture decisions and their consequences. Migrating a smart-pet-device company off Objective-C incrementally instead of rewriting, and what +30% on perf and UX metrics cost in dual-stack interim time. Standing up a modular SwiftUI codebase so two Williams-Sonoma apps could share it. Splitting a monolith while the team keeps shipping every sprint.
The calls I got wrong. Every post that has a reflection section has one because I made the mistake first. I set the metric-of-record after deploying agents instead of before. I underestimated the dual-stack interim. I refactored a shared module out after the second app diverged rather than extracting it from commit one.
What you won’t find: AI news commentary, career advice, or takes on tools I have not run in production.
Background, briefly
Ten-plus years in iOS, five-plus leading engineers. Shipped App Store apps for Stanley by Neiman Marcus, eBay, Pottery Barn Kids and Teen, Whistle, and AE + Aerie. Kyiv Polytechnic grad. Ukrainian, based in Valencia.
The through-line across all of it is diagnosis before building: on the Stanley engagement the win was telling a stuck React Native team what not to rebuild, which unblocked a launch a native rewrite would have delayed by quarters.
Working together
I take on a small number of advisory engagements: architecture reviews, AI-in-the-review-loop setups, and second opinions on migration plans teams are about to commit a year to. What that looks like in practice.
- Book 30 minutes: calendly.com/oleh-veheria
- Email: oleh@veheria.tech
- LinkedIn: linkedin.com/in/oleh-veheria
If you are about to put agents into your own review loop, the one question worth settling first is which part of review stays human. What is your answer?