Tool map: 5 roles, analysis of services and comparison table

Map of tools and service roles

A beginner does not need to know “all the services in the world.” Enough to understand 5 roles. Below are the roles, and then a detailed analysis of each key service.

Role 1: “Think and formulate”

This is the “brain” and the “editor”: formulate the idea, requirements, texts, plan, screen structure.

Examples: ChatGPT, Claude, Gemini - any LLM-model in chat.

⚠️ Practical rookie mistake: immediately ask “make an application” without explaining what exactly and how to check it.

Role 2: “Help with code”

Assistants in the editor/terminal: they prompt, add, explain, fix.

Examples: GitHub Copilot, Cursor, Windsurf - AI assistants and pilots in IDE.

Role 3: “Agent - do a task for me within the project”

The agent sees the project, changes several files, can run commands/tests, asks for permission to perform dangerous actions.

Examples: Claude Code, Cursor Agent, Windsurf Cascade - agents inside the IDE/terminal.

Role 4: “Build a turnkey application”

Platforms that close the path from “description” to “ready URL”, often with backend-opportunities and publication.

Examples: Replit (Agent + Plan Mode), Lovable (full-stack in steps).

Role 5: “Automation”

You connect services: “an application has arrived → write to the table → send a notification → start AI processing.”

Example: n8n — workflow automation tool.

Bonus: “Deployment and Environments”

Where the project lives and how it is being rolled out: preview i production.

Example: Vercel – preview-deploy to the /PR, production branch when merging in main.

Detailed analysis of services

OpenAI: ChatGPT, Codex and OpenAI API

Entry threshold: low/medium

• What is this. ChatGPT - “universal brain”. Codex is a coding agent that works in the project: reads files, edits, runs commands. API - connecting models directly.
• Strengths. gpt-5.3-codex: context up to 400k, customizable reasoning effort.
• Limitations. Price depends on tokens. An agent can break a project without checkpoints. An external deployment layer is needed.

GitHub Copilot

Entry threshold: low/medium

• What is this. Assistant in IDE: autocompletion, chat, coding agent (branch → commits → PR → review).
• Strengths. Agent in the usual PR review. Speeds up minor edits, refactorings, tests.
• Limitations. Limits requests. Without knowledge of branches/PR, “agent through PR” is an extra layer.

Anthropic: Claude and Claude Code

Entry threshold: medium

• What is this. Claude Code - agent: reads the code base, edits files, runs commands.
• Strengths. Permissions and checkpoints for rollback. Project memory via CLAUDE.md. Context engineering.
• Limitations. Price/usage limits. You need a terminal and discipline.

Cursor: AI-first IDE

Entry threshold: medium

• What is this. Fork of VS Code with Agent Mode: plans, edits several files, runs tests.
• Strengths. Supports different models. Inline editing. Familiar VS Code interface.
• Limitations. Desktop application. Request limits. The agent can “leave” without checkpoints.

Windsurf: AI IDE with Cascade

Entry threshold: medium

• What is this. AI IDE with Cascade is a flow agent that sees the entire project and tracks actions in real time.
• Strengths. Multi-step tasks. Contextual awareness of edits. Several models.
• Limitations. New product, smaller ecosystem. Flow credits limits.

Google Antigravity

Entry threshold: medium

• What is this. Agent-first IDE from Google. Cross-surface agents: editor + terminal + browser simultaneously.
• Strengths. Agent Manager for multiple agents. Task-based approach. Now free.
• Limitations. Only Linux for now. A new product, an ecosystem is being formed.

Replit Agent

Entry threshold: low

• What is this. Cloud environment: code + launch + publish. Replit Agent + Plan Mode for iterations.
• Strengths. Minimum local configuration. Plan Mode helps you avoid turning your project into a mess.
• Limitations. Credit model. As you grow, you'll want to push the code to GitHub.

Lovable: prompt → full-stack web

Entry threshold: low

• What is this. Full-stack AI platform for creating and deploying applications “as you say.”
• Strengths. Live URL with publication. Native integration with Supabase. Key protection. Export to GitHub.
• Limitations. Credit model. Complex systems require education on the backend.

n8n: automation and AI agents

Entry threshold: low/medium

• What is this. Automation platform with AI Agent node for actions via API.
• Strengths. “Glue” between services: webhooks, HTTP, triggers. AI nodes in human language.
• Limitations. The complexity increases with the scale of the workflow. Execution limits.

Vercel

Entry threshold: low

• What is this. Deployment platform: each deployment provides a unique URL for preview.
• Strengths. Preview-environments (PR → preview). Manage env vars. Deploy = build result.
• Limitations. Usage-based expenses (monitor billing).

Supabase: “backend in a box”

Entry threshold: medium

• What is this. Postgres + auth + APIs + edge functions + realtime.
• Strengths. Postgres and "instant API". Documentation with examples of billing calculations.
• Limitations. You need to master tables, authorization, access policies.

Firebase

Entry threshold: medium

• What is this. BaaS platform with Spark (free) and Blaze (pay-as-you-go).
• Strengths. Clear model: two plans, calculators and quotas.
• Limitations. The cost is “spread out” across services. Limits and monitoring are needed.

Git: version control system

Entry threshold: medium

• What is this. Git is a distributed version control system. Records every change and allows you to roll back to any version.
• Why is it critical in vibe coding. The AI agent can rewrite dozens of files in a minute. Without Git, rolling back a bad change is a pain.

Key Concepts:

• Repository —project folder with change tracking.
• Commit — “snapshot” of changes. How to save in game.
• Branch —parallel development line.
• Merge - merging changes from one branch to another.
• Pull Request (PR) —request to merge with review.

Minimum set of commands: git init, git add ., git commit -m "...", git push, git pull

• Entry threshold. Intermediate: basic commands in a day, but merge conflicts require practice.

GitHub

Entry threshold: low

• What is this. The largest platform for Git repositories and collaboration. De facto standard.
• Strengths. Issues, PRs, Actions for CI/CD, Copilot, Pages, open-source ecosystem.
• Limitations. CI/CD: 2,000 min/month on the free plan.

GitLab

Entry threshold: medium

• What is this. DevOps full-cycle platform: Git + CI/CD + container registry + security scanning.
• Strengths. Everything out of the box. Self-hosted. GitLab Duo - AI assistant.
• Limitations. The interface is more complicated than GitHub. CI/CD: 400 min/month on the free plan.

Unified comparative table of services

Relevance: March 2026. Prices/limits - from official documents.

Choose a tool that suits the task, not the “most hyped” one. If you need a quick web result, use app builder. If you need control - IDE/agent + Git. If you need a business process - automation.