Building with AI as the team.
Strategy, product, engineering, finance, marketing — one founder, AI as the team. Field notes from running JourneyBay, Kronos, and a one-person stack.
Flagship launches
What I’m building
Two products that show how I run AI as the team.
Kronos Agent OS: a self-hosted runtime for durable AI agents
An evidence-based tour of KAOS v0.3.0: durable turns, layered memory, MCP tools, governance, behavior evals, portable agent state, and safe defaults.
JourneyBay 2.0: An AI Travel Engine That Holds the Entire Trip
JourneyBay 2.0 is live: an AI Travel Engine that edits your trip, not just answers questions. A trip from one prompt, booking import, visa context, and MCP access.
Featured reading
Where to start
Three pieces that capture how I think about AI-augmented building.
Context Engineering for LLM Agents: A Practical Guide
Build reliable context for LLM agents: select evidence, separate instructions from data, manage tools and memory, compact safely, and test each model call.
Claude Code Guide: Setup, CLAUDE.md, Skills, MCP, CI
AI coding for Flutter: a verification-first workflow
Latest
Fresh from the workbench.
GEO in Practice: 3,500 Copilot Citations and 25 Bing Clicks
Marketing Cases & PracticeA generative engine optimization case study: what Copilot cites, for which queries, and why citations grow but clicks don't. 3 months of Bing AI data.
Prompt Caching: Cached Input 10-40x Cheaper (Claude, OpenAI, Gemini)
Tutorials AI OpsHow prompt caching works on Claude, OpenAI, Gemini and DeepSeek: price multipliers, TTL, a break-even formula, OpenRouter measurements and why cache misses happen.
LiteLLM Proxy Guide: Self-Hosted LLM Gateway for Teams
Tool Reviews TutorialsSet up LiteLLM Proxy: config.yaml, virtual keys with budgets, Ollama-to-cloud fallback, Claude Code and OpenAI SDK clients, 2026 CVEs and hardening.
Context Rot: Why Long Context Degrades LLM Answers
Engineering AI OpsWhat context rot is, why LLM answers degrade long before the context window fills, what Chroma, NoLiMa and RULER found, a Claude Haiku 5.5 test and fixes.
OpenCode Guide: Open-Source Alternative to Claude Code
Tool ReviewsWhat OpenCode is, how to install it, connect cloud and local models via Ollama, and set up agents, permissions, and MCP. How it compares to Claude Code.
llms.txt: Do You Need It and How to Build It Right
Tutorials MarketingWhat llms.txt is, who actually reads it (Ahrefs data, Google's position), how to generate it and serve .md pages, and what it does and does not do for GEO.
By topic
Four pillars I write about most.
Engineering
Software engineering practices, architecture, testing
- Context Rot: Why Long Context Degrades LLM Answers
- How to Test an MCP Server: Contracts, Failures, and CI
- Terraform Modules with AI: IaC for Startups Without a DevOps Engineer
Product
Product management, research, specs, prioritization
- Edge Cases in PRDs with Claude: 23 Missed Scenarios in One Prompt
- Usability Test Script with AI: A Step-by-Step Guide
- Design System Prompt Library: Getting Consistent UI from AI
AI Ops
Running AI systems in production
- Prompt Caching: Cached Input 10-40x Cheaper (Claude, OpenAI, Gemini)
- Context Rot: Why Long Context Degrades LLM Answers
- Claude Code Subagents: When They Save Context and When They Hurt
Strategy
Business models, market sizing, competitive analysis
- Living CI Document: Competitive Intelligence That Updates Itself
- Competitive Intelligence Report in 30 Minutes: Prompts and Template
- Feature Matrix Template: Competitor Screenshots to Insights with AI
About the author
Roman Belov — technical founder. 10+ years in product (Yandex, OZON, VK), now full-time on AI-augmented building. Writes weekly about engineering, product, and operations with AI as the team.
More about me →