Vahe Khumaryan — Notes https://vahe.space/notes/ Short notes on building systems, scaling support, and shipping AI-assisted software. en Sat, 13 Jun 2026 09:00:00 +0400 The unglamorous tool usually wins https://vahe.space/notes/unglamorous-tool/ https://vahe.space/notes/unglamorous-tool/ Sat, 13 Jun 2026 09:00:00 +0400 Notes on internal tooling, and the terabytes of video nobody could find. — The internal tools that matter most are almost never the impressive ones. They're the boring utilities that quietly remove a daily papercut for an entire team. Fluent in both the codebase and the org chart https://vahe.space/notes/codebase-and-org-chart/ https://vahe.space/notes/codebase-and-org-chart/ Fri, 12 Jun 2026 09:00:00 +0400 Why the most useful people sit on the seam between building and leading. — Most careers push you to pick a lane: go deep technical, or move into management. The two get treated as a fork in the road. I've spent mine refusing to choose, and it's been the most useful decision I made. Agents that do work, not demos https://vahe.space/notes/agents-that-do-work/ https://vahe.space/notes/agents-that-do-work/ Wed, 10 Jun 2026 09:00:00 +0400 The gap between an impressive AI demo and one you'd trust with real operations. — Every week there's a new agent demo that looks like magic. Most of them would not survive a Tuesday in a real operation. The gap between the two isn't model quality — it's everything wrapped around the model. Support that scales is a system, not a headcount https://vahe.space/notes/support-is-a-system/ https://vahe.space/notes/support-is-a-system/ Mon, 08 Jun 2026 09:00:00 +0400 What 86 million users and a from-scratch help desk taught me. — When support volume grows, the instinct is to add people. People do help. But headcount is the lever you reach for when you don't have a system — and it stops working long before the users stop arriving. When to build the tool instead of buying it https://vahe.space/notes/build-vs-buy/ https://vahe.space/notes/build-vs-buy/ Fri, 05 Jun 2026 09:00:00 +0400 A paid platform isn't always the cheap option. — The default move when a team hits a process problem is to buy a SaaS for it. Usually that's the right call. Occasionally it's an expensive mistake hiding behind a reasonable-looking invoice.