Notes
This section presents documents that do not claim to be full-fledged books, as they were created by me with AI on the basis of existing documentation and articles publicly available on the internet. They serve for quickly reviewing or acquiring the knowledge you are missing about technologies and the principles of designing applications and sites. These are more like summaries than full-fledged books.
Tools
Top AI models - orchestrator and agents
A reference on the orchestrator-and-subagents pattern: how it is built in the leading AI coding environments, where the config and the model behind each one live, and which meta-prompt sets it up for a specific project - grounded in official sources.
Tools
Devin Desktop: From Windsurf to a System of Agents
The former Windsurf became Devin Desktop - not just an editor but several execution environments around one agent: the local IDE, Devin Local, Devin Cloud, Cascade and the shared ACP protocol. How to pick a surface for a task, what limits the agent in each, and how to prove a result before it reaches the repository.
Tools
Cursor: Where Your Agent Runs
The same command does different things in the editor, the terminal and the cloud, because those are different environments with different rights. How to pick a surface for a task, what limits the agent in each of them, and how to verify a result before it reaches the repository.
Tools
Codex: Controlled Autonomy
Autonomy is useful exactly as far as it is controllable. How reasoning depth and the cost of a run are configured, how approvals differ from the sandbox, what an execution policy is for, and why the same command behaves differently in the terminal, the editor and the cloud.
Tools
Claude Code: The Agent Under Supervision
An agent does exactly what it is allowed to do - the question is who sets those limits and where. The working loop and permissions, project memory that outlives a session, skills and subagents, and how all of it behaves in continuous integration with nobody at the screen.
Testing
E2E and Playwright: Tests That Outlive Refactoring
An end-to-end test breaks not because the app broke but because it was written against the markup. Locators that survive a redesign, waiting without sleep, test data and authentication, and how to keep such a suite for years rather than until the first green run.
Testing
Vitest: Testing in the Vite World
When the build runs on Vite, testing follows its own logic: the same config, the same transforms, the same aliases. What that implies, where Vitest departs from Jest habits, and what Browser Mode gives you once jsdom stops resembling a browser.
Testing
Jest: Isolation, Mocks and a Suite That Lasts
Tests get expensive when they fail for the wrong reasons. Isolation that keeps a neighbouring test from breaking yours, mocks at the boundaries rather than inside the logic, snapshots that do not rot, and a suite that survives a refactor.
Frontend
Accessibility, SEO and i18n
Three requirements usually handled separately and reworked three times over. They share one root: meaningful markup and honest text. One architecture that answers a screen-reader user, a search crawler and a translator alike.
Frontend
PWA and Offline-First
An app that survives the subway, the lift and bad station Wi-Fi. What the service worker does when the network lies, which data to keep locally, how to reconcile it with the server later, and why installing on a device raises the bar for quality.