Live Nov 18 · Cohort 3Nov 18, 19:00 MSK — free hands-on AI agents live workshop · Live Cohort 3 starts Nov 23Join Nov 18 Live →Cohort 3 (Nov 23)

"Tochki nad II" Author Replaces Notion and CRM with Agents: 80+ Skills Instead of Services

The author of the "Tochki nad II" channel shared what he uses every day. At the center of everything is a window with an agent that has access to all projects and repositories: Claude Code and Codex on main subscriptions. Everything else is scaffolding around that window. The main effect: more than 80 custom skills replaced presentation builders, task trackers, CRM and Notion.

AI-processed from TG: @TochkiNadAI (Точки над ИИ); edited by Hamidun News
"Tochki nad II" Author Replaces Notion and CRM with Agents: 80+ Skills Instead of Services
Source: TG: @TochkiNadAI (Точки над ИИ). Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTG: @TochkiNadAI (Точки над ИИ) ↗
Editor-in-Chief
Key metric #1Skills handle contracts, invoices, SEO audits and call analysis
Key metric #2A single OpenRouter key covers almost any model
Key metric #3The agent runs the Supabase backend directly through MCP

While everyone argues about which model is best, the author of the "Tochki nad II" channel showed what a workday built around an agent looks like. It is not a list of favorite apps but the design of a workspace: one agent window (IDE, terminal or native app) with access to all projects and repositories. Everything else, in his words, is scaffolding around that window.

What Is at the Center: Agents and Subscriptions

The foundation is Claude Code and Codex on main subscriptions. He uses them both in apps and in the iTerm terminal. Orca is needed to run several agents in parallel in one window on his own subscriptions. Gemini and Cursor are also paid for, but the author admits he keeps them mostly for demonstrations at the camp. Sometimes he opens Dots from OpenAI.

Voice and browser are built around the agents. MacWhisper transcribes videos and calls and puts the result into Second Brain on its own. For dictation there is "Klava ne vinovata" ("The Keyboard Is Not to Blame"), a custom build of Handy made by a channel subscriber. The Comet browser hosts agent extensions from OpenAI and Anthropic.

The stack is not monogamous: two main subscriptions, plus an aggregator, plus separate keys for projects. This is normal practice for those who work with agents every day and do not want to depend on a single vendor.

Code, Infrastructure and APIs

All code lives on GitHub. Telegram bots run on Railway. The backend is built on Supabase via MCP, and sometimes Neon is used for databases. Demos are deployed to Netlify and Vercel, Mux serves video, and clips are recorded in Screen Studio.

For APIs the setup is this: OpenRouter with one key for almost any model, plus separate OpenAI, Gemini and Kimi keys for specific projects. When the agent needs to pull an entire website, Firecrawl is connected.

The common trait of this set is that almost every service can be controlled from the agent via API or MCP. The agent does not just write code: it deploys, connects the database and fetches the data itself.

Skills Instead of Services

The most interesting part of the post is the skills. There are already more than 80, and many replace entire services. Examples include:

  • camp-style presentations;
  • carousels and posters;
  • a broadcast intro and timecodes for a lesson;
  • translation and layout of PDF guides;
  • a contract, invoice and completion act for camp participants;
  • call analysis in the CRM inside Second Brain;
  • SEO and GEO site audit;
  • diagnostics for a slow Mac.

The result, according to the author: he left presentation builders, task trackers, CRM and Notion, and in their place came personal and team micro-services for a host of small tasks.

Put the items together and a clear chain emerges. A call with a participant is recorded and transcribed by MacWhisper, the text lands in Second Brain, one skill analyzes the call in the CRM, and another prepares the contract, invoice and act. Previously this took several services and manual copying between them.

Limitations and What to Take Away

A caveat up front: this is a list-style post, not research. There are no figures on time or money saved, and subscription costs are not named. The stack is tailored to one person who creates content, training and camps, and it will not fit a developer or product manager one to one. An agent with access to all repositories is also a question of permissions and security, which the post does not address. Finally, 80 skills need maintenance, or they will go stale and start making mistakes.

The practical takeaway can be put without copying the list:

  • write down the recurring tasks for which you pay a service only for a template or form;
  • turn the most frequent one into an agent skill and compare the result with the service;
  • connect infrastructure via MCP or API so the agent can take a task to completion;
  • set up one aggregator key like OpenRouter and separate keys only where you need control over a project;
  • decide in advance which repositories and data the agent gets access to.

The main idea of the post is simple: value today lies not in a set of subscriptions but in how much routine you have managed to turn into skills and remove from third-party services.

Story

Key Questions Answered

What Is at the Center: Agents and Subscriptions?

The foundation is Claude Code and Codex on main subscriptions. He uses them both in apps and in the iTerm terminal. Orca is needed to run several agents in parallel in one window on his own subscriptions. Gemini and Cursor are also paid for, but the author admits he keeps them mostly for demonstrations at the camp. Sometimes he opens Dots from OpenAI. Voice and browser are built around the agents. MacWhisper transcribes videos and calls and puts the result into Second Brain on its own.

What are the key details on Code, Infrastructure and APIs?

All code lives on GitHub. Telegram bots run on Railway. The backend is built on Supabase via MCP, and sometimes Neon is used for databases. Demos are deployed to Netlify and Vercel, Mux serves video, and clips are recorded in Screen Studio. For APIs the setup is this: OpenRouter with one key for almost any model, plus separate OpenAI, Gemini and Kimi keys for specific projects. When the agent needs to pull an entire website, Firecrawl is connected.

What are the key details on Skills Instead of Services?

The most interesting part of the post is the skills. There are already more than 80, and many replace entire services. Examples include: camp-style presentations; carousels and posters; a broadcast intro and timecodes for a lesson; translation and layout of PDF guides; a contract, invoice and completion act for camp participants; call analysis in the CRM inside Second Brain; SEO and GEO site audit; diagnostics for a slow Mac.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…