English notes and imported posts about AI agents, Claude Code, data and Telegram.
Материалы
- 183 AI Bots on Telegram with 44M Users: What the Data Shows — Research into AI agents on Telegram: 183 bots with 44.3M combined MAU. BotFather doubled to 7.3M. Telegram is the default AI product interface.
- 21 Questions That Expose Gaps in Your AI Agent's Knowledge — A curated list of uncomfortable questions to ask Claude Code, Codex, or any AI coding agent. Each answer reveals blind spots in how the agent understands you.
- AI Tools I Actually Tried in March 2026: Honest Review — Hands-on review of 5 AI tools from March 2026: paperclip task tracker, gstack YC prompts, conductor.build IDE, x402 crypto payments, and agent-browser setup.
- Always-On AI Agent: Running Claude Code 24/7 on a Server — How I run an always-on AI agent on a Hetzner VPS. Architecture, Docker setup, Telegram bridge, memory system, and real costs.
- Telegram MCP Server for Claude Code: Telethon, MTProto and Real Workflows — How to build a Telegram MCP server for Claude Code with Telethon: tools, auth, session strings, async traps, security boundaries and production workflows.
- Claude Code /compact: Context Compression, Compaction Prompts, What Survives — How Claude Code /compact works: context compression, what survives compaction, prompt caching cost, PreCompact/PostCompact hooks, and practical recovery tips.
- Decentralized AI Compute: Cocoon vs Bittensor vs Gonka Compared — Comparing three decentralized AI compute protocols: Cocoon (TON), Bittensor (TAO), and Gonka (GNK). Architecture, tokenomics, validation, and real-world performance.
- My Claude Code Setup: MCP Servers, Hooks, Skills and Agents (2026) — Claude Code setup 2026: MCP servers, MCP hooks, skills, Windows notes and the minimal working setup I would configure before adding agents.
- Claude Code + Codex Dual Review: How Two AI Agents Cross-Check Each Other's Plans — How to set up cross-validation between Claude Code and OpenAI Codex via MCP. Step-by-step Dual Review protocol with prompt examples.
- Telegram Ads Keyword Analysis: 61,000 Search Queries from 3,100+ Advertisers — Analysis of 61K search queries from Telegram Ads advertisers. Top niches: tarot, dating, VPN. 3,100+ advertisers, 10,600 unique keywords.
- Fake Claude Code Website: How I Found a Phishing Site in Google Ads — Phishing site impersonating Claude Code found in Google Ads: copied design, base64-encoded install script, malicious binary. How to spot the fake.
- How I Automated TON Blockchain Data Analysis with an AI Agent — Habr Article — Story of automating TON Blockchain data analysis by building ton-analyst skill for Claude Code. Full article on Habr with examples and code.
- BotFather MAU Up 43% in 6 Weeks: Telegram as a Platform for AI Agents — BotFather MAU grew from 3.5M to 5M in 6 weeks (+43%). Telegram is a convenient interface for AI agents. Stats on python-telegram-bot, grammY, aiogram.
- ton-analyst: Open Source AI Skill for TON Blockchain Analysis via Dune Analytics — Open source ton-analyst skill for Claude Code: automated TON Blockchain data analysis via Dune Analytics. Installation and usage guide.
- The Singularity Already Happened — Just Not Where You Think: Analysis of 5 AI Metrics — Analysis of 5 AI progress metrics: only academic papers show hyperbolic growth. Everything else is linear. Singularity = attention, not capabilities.
- Web Scraping AI Agents: What Actually Works in 2026 — Is web scraping dead in 2026? Not exactly. Use AI agents to choose API/XHR, Playwright/browser flows or LLM extraction without hallucinated selectors.
- How to Write Claude Code Skills Properly: 7 Common Mistakes to Avoid — 7 mistakes that break Claude Code skills: no trigger, generic names, empty sections, verbose triggers. Checklist for a good skill file.
- My AI Dev Tools in 2026: What I Actually Use Daily — Ghostty, Descript, Coolify, and 4 other tools in my daily dev workflow. How they actually fit together after months of use.
- 5 Prompts That Improved My AI Agent Workflow: Tested Hacks for Claude Code — Five battle-tested prompts for Claude Code: ask clarifying questions, don't stop till done, show-don't-tell with playground, multi-agent hiring, memory from mistakes.
- dania.zip: packaging yourself for AI agents — How I created an exportable personality spec for AI agents. File structure, tone extraction, and real use cases for a digital twin.
- Obsidian Second Brain with Claude Code: My Agent-Readable Vault Setup — How I use Obsidian, Claude Code, CLAUDE.md and voice notes as an agent-readable second brain: folder structure, rules, workflows and failure modes.
- Where to Find SaaS Ideas: acquire.com Reverse Engineering with AI Validation — How to find profitable SaaS ideas: parse acquire.com listings, reverse-engineer competitor marketing, validate with AI. Step-by-step checklist from idea to MVP.
- Claude Code workflow: how I code with AI in 2026 — My workflow with Claude Code: plan mode, spec-driven development, and why AI agents replaced autocomplete. Real examples and costs.
- Tips from Claude Code Creator: Git Worktrees, Self-Improving CLAUDE.md, and Code Review — Boris from Anthropic shared Claude Code tips: git worktrees for 3-5 parallel sessions, self-improving CLAUDE.md, Claude as strict reviewer, /techdebt skill.