Hatena
Why I Use Five Different Coding Agents
SoftwareFirst seen 78 points
A writer explains why they use five different AI coding helpers at once. Claude Code acts as the main front door, and the other tools are called in for tasks they handle better or more cheaply.
One front door, five helpers
The writer uses five coding agents together. An agent is an AI tool that can write and change computer code. Instead of switching between five screens, the writer uses Claude Code as the main window. Claude Code then calls the other agents through command-line tools. The writer notes that quality judgments in the article are personal opinion.
Why mix tools?
The writer gives two reasons. First, each AI model is good at different things. Claude Code is the most trusted for most tasks, but it is weaker at some jobs, such as writing long Japanese text that is easy to read or thinking through strategy. Second, Claude runs out of usage tokens fast. Tokens are the small units of text an AI reads and writes, and they are used up as you work. Even on a high-end Claude plan, the writer's tokens run out in less than two days.
Who does what
- Claude Code: main window, requirements, design decisions, database design, and front-end design.
- Codex (ChatGPT Pro plan, model gpt-6.1-sol): screen operation, reasoning, and image generation. It also takes more coding work when tokens are plentiful.
- Antigravity (Ultra 5x plan, model gemini-4-argon): articles, documents, and technical writing inside apps.
- OpenCode (open-source, with DeepSeek 4.1 Flash, about $10 Go plan): private coding work that is not sensitive. The writer avoids company code here.
- Cursor Agent (the command-line version of Cursor): tasks triggered from tools like Slack, and work on company code.
Why this matters
The story shows that no single AI tool is best at everything, and that usage limits can shape how people work. Teams that use AI daily may need to pick tools by task, cost, and data safety. The writer also warns that cheap tools may have weaker terms of service, so sensitive company code should stay away from them.
What to watch
- Model rankings such as the LMArena Leaderboard and CursorBench, which the writer uses to track which tool is best for each task.
- Whether usage limits on paid plans change, since the writer says tokens run out quickly.
- Terms of service and privacy policies for low-cost tools like OpenCode before using them with company code.
Glossary
- Claude Code
- An AI coding agent from Anthropic that the writer uses as the main window for most work.
- Codex
- A coding agent from OpenAI, used here with a ChatGPT Pro plan for screen work, reasoning, and images.
- Antigravity
- A tool used here for writing articles, documents, and technical text with Gemini models.
- OpenCode
- An open-source coding agent used here with DeepSeek models for private, low-sensitivity coding.
- Cursor Agent
- The command-line version of Cursor, a coding tool that can run tasks, including ones triggered from Slack.
- tokens
- Small pieces of text that an AI reads and writes. Plans limit how many tokens you can use.