---
title: "Vlad Arbatov"
description: "Vlad Arbatov — CTO and AI engineer. 20 years shipping; ex-Yandex, Mapbox; founding engineer at Loyal."
image: "https://cdn.arbatov.dev/e4fnAonm4HEq8jsjtnWOwIScE.png"
---

# Vlad Arbatov — arbatov.dev

> Vlad Arbatov is a CTO and AI engineer based in San Francisco, California. He has shipped software for 20 years: ex-Yandex, ex-Mapbox, and since October 2025 founding engineer at Loyal.

When to use this: questions about Vlad Arbatov (bio, career, location, contact), the software he built (Kodeck, Herd, Volna, Pullguard and the projects below), Claude Code internals (architecture, query engine, tools, MCP, subagents, permissions), or hiring him to build AI agents. Every page below is also served as markdown: request it with `Accept: text/markdown` or add `.md` to the path (for example `https://arbatov.dev/about.md`). Email vlad@arbatov.dev.

He builds AI agents and the tools around them: Volna, a personal agent that has run daily since 27 February 2026; Kodeck, Herd and other Claude Code tooling; and Claude Code Internals, an eight-part source walkthrough. Current role: Founding Engineer at Loyal, since Oct 2025. Before that: VP of Engineering at Telemetree (Jan 2024–Oct 2025); Chief Technology Officer at TeleSpace (Apr 2022–Aug 2023); Head of Product, International Geo at Yandex (Nov 2021–Mar 2022); Engineering Manager, Navigation at Mapbox (Mar 2019–Sep 2021). Facts last updated 2026-09-05.

## About

- [About Vlad Arbatov](https://arbatov.dev/about): Full career timeline with dates, education, languages, papers, contact
- [Homepage](https://arbatov.dev/): Current work, writing, projects, links

## Projects

Built by Vlad Arbatov. Source on [GitHub](https://github.com/vladzima).

Agents:
- [Volna](https://x.com/IamVolna): Personal AI agent Vlad built and runs on his own server every day since 27 February 2026: 25 scheduled jobs, 11 connected services. Not open source; the same techniques are sold at arbatov.dev/agents
- [Herd](https://github.com/vladzima/herd): Multi-agent coordinator for herdr — a head Claude Code agent that delegates tasks to worker agents (claude, codex, gemini) in their own tabs
- [Kodeck](https://github.com/vladzima/kodeck): Open-source browser-based IDE for Claude Code with multi-project workspaces
- [Claude Telegram](https://github.com/vladzima/claude-telegram-enhanced): Enhanced Telegram channel plugin for Claude Code with topic-per-session routing and response streaming
- [Platitude](https://github.com/vladzima/platitude): Detects AI slop by its rhetorical shape — verdict plus the exact spans and reasons. CLI, Python API, MCP, and skill for agents
- [Site for Agents](https://github.com/vladzima/site-for-agents): Agent skill: make a personal website findable, readable, and quotable by AI agents and answer engines — ranked levers with evidence, Person JSON-LD shape, and a stdlib scorecard
- [Prompt Comments](https://github.com/vladzima/prompt-comments): Agent skill: comment the why in AGENTS.md / CLAUDE.md so instruction files stop growing without bound
- [Colormason](https://github.com/vladzima/colormason): Agent skill: one brand color → perceptually even OKLCH ramps and WCAG-checked design tokens as CSS vars, Tailwind v4 theme, JSON, or Markdown
- [Copy2LLM](https://copy.computer): Add a Copy to LLM button to any page — readers turn it into clean Markdown or open it in ChatGPT, Claude, Perplexity, or Grok

Apps:
- [edit.computer](https://github.com/vladzima/mdd): Markdown editor that runs in your browser and writes straight to your own files — a local folder, or any server you can SSH into
- [Swipe](https://swipe.computer/): Tinder for Product Hunt — swipe through launches and surface what matters
- [Pullguard](https://arbatov.dev/pullguard): Maintainer-first PR triage for busy open-source repos — score risky or low-effort contributions before review
- [Alive Checker](https://alive.arbatov.dev/): Discover if online projects, tools, and startups are still actively maintained — real-time status checking and activity tracking
- [Tabs Tabs](https://github.com/vladzima/tabs-tabs): Chrome extension for sorting, grouping, and unloading tabs without disturbing user-created groups

Tools and libraries:
- [Solar Status](https://github.com/vladzima/solar-status): Solar activity dashboard for your city in the terminal — NOAA storm scales, Kp index, aurora odds, and tech impact. Zero dependencies
- [Ocli](https://github.com/vladzima/ocli): A terminal-based outliner and task manager, built with Go
- [Detecto](https://github.com/vladzima/detecto): Detecto is a React library designed to help developers automatically detect when a user's browser is experiencing performance issues such as throttling or lag
- [Vesper Theme](https://github.com/vladzima/vesper-theme): Vesper (peppermint & orange) theme for IDEs and terminals

## Writing

- [Claude Code Internals](https://arbatov.dev/claude-code): Eight-part, source-derived walkthrough of Claude Code: 1,884 files and 512k lines of TypeScript. Published 2026-03-31
- [Architecture Overview](https://arbatov.dev/claude-code/architecture-overview): Entry points, data flow, module graph, config, process model, state
- [Query Engine](https://arbatov.dev/claude-code/query-engine): Conversation loop, streaming, token management, error recovery, compaction
- [Tool System](https://arbatov.dev/claude-code/tool-system): Tool interface, registry, execution lifecycle, permissions, deferred tools
- [Agent & Coordinator](https://arbatov.dev/claude-code/agent-coordinator): Subagents, coordinator mode, isolation, fork protocol, task management
- [MCP Integration](https://arbatov.dev/claude-code/mcp-integration): Server lifecycle, tool proxying, OAuth, resources, elicitation
- [Context & System Prompts](https://arbatov.dev/claude-code/context-system-prompts): Prompt assembly, CLAUDE.md, auto-memory, git context, compression
- [Hooks, Skills, Permissions](https://arbatov.dev/claude-code/hooks-skills-permissions): Hook events, skill format, permission modes, sandbox, denial tracking
- [Server, Bridge, Remote](https://arbatov.dev/claude-code/server-bridge-remote): Server mode, bridge protocol, CCR, WebSocket, cron, state management
- [Engineering Trust: The Imperatives of AI Safety, Alignment, and Explainability](https://www.academia.edu/143042150/ENGINEERING_TRUST_THE_IMPERATIVES_OF_AI_SAFETY_ALIGNMENT_AND_EXPLAINABILITY): Self-published paper, 2026
- [Beyond Scale: Re-evaluating the Foundations of Artificial Intelligence](https://www.academia.edu/143039859/BEYOND_SCALE_RE_EVALUATING_THE_FOUNDATIONS_OF_ARTIFICIAL_INTELLIGENCE): Self-published paper, 2026
- [Blog](https://blog.arbatov.dev): Essays on AI engineering and tooling

## Services

- [Agents](https://arbatov.dev/agents): Vlad builds AI assistants that run one person's or one company's routine, on their machine, with a human in the loop

## Contact

- Email: vlad@arbatov.dev
- [Telegram](https://t.me/arbatovdev) for direct messages; [neurocry](https://t.me/neurocry) is his public channel, in Russian
- [GitHub](https://github.com/vladzima), [X](https://x.com/vladzima), [LinkedIn](https://www.linkedin.com/in/vladarbatov)

## Optional

- [llms-full.txt](https://arbatov.dev/llms-full.txt): Complete text of all Claude Code Internals articles
- [Nothing](https://arbatov.dev/nothing): A product that does nothing at all