Introducing Factory.ai & Droid: Beyond Claude and Codex?

👋 Introduction

I would like to introduce Factory.ai and Droid, which is currently my most recommended AI coding platform.
This AI has been a hot topic recently, achieving scores that surpass Claude Code and others on the Terminal-Bench benchmark.
Droid: The #1 Software Development Agent on Terminal-Bench

The reason I intentionally call it an "AI coding platform" rather than an "AI coding agent" is because I feel it goes beyond being a mere agent, allowing for autonomous development while collaborating with other applications.

While you should refer to the official documentation for more details, this article will briefly introduce the features of Factory.ai and Droid.

🏭 What is Factory?

Factory.ai is an AI software development platform that, to put it bluntly, feels like "Devin + Claude Code + GitHub Copilot + Spec-driven".
While development is centered around the Droid CLI tool, it offers the following distinctive features that set it apart from other AI coding agents:

• You can execute tasks by entering prompts on a web app.
• You can choose from multiple leading AI models such as Open AI GPT-5 and Claude Sonnet.
• You can integrate with and issue instructions through various tools like GitHub, GitLab, Jira, Slack, and Linear.

🚀 How to Get Started

You can start using it right away just by registering an account on Factory.ai.
There is a one-month free trial, so I encourage you to give it a try.

Next, install the Droid CLI tool and Factory Bridge, which links the web and CLI. For details, please refer to the Quickstart guide.

• Quickstart | Factory Documentation

💰 Pricing Structure

Now, regarding the pricing structure you might be curious about, it is a monthly subscription based on the number of tokens. (As of October 29, 2025)

• Pricing Plan
• Pricing
  

The Pro plan is $20 per month and includes up to 20 million tokens. This means you can use models from both companies at the same price as Claude Pro or ChatGPT Plus.
Additionally, since the token consumption is designed to vary by model, you can accomplish a significant amount of work for $20 depending on your choice of model.
In particular, GPT-5 and GPT-5-Codex have a token consumption rate of 0.5x, making them very economical to use actively.

• Model Pricing
  

If you exceed the tokens included in your plan, you can purchase additional tokens at $2.7 per 1M tokens for the excess.

By the way, I also use Claude Code with a Claude Pro plan, and my impression is that
if you use it to the full extent of the 5-hour limit, Claude Code allows for more work at the same price compared to Factory.
Heavy users might find Factory less cost-efficient.

You can also use your own API keys (BYOK), but please note that this is limited to the Droid CLI.

🤖 Droid

Droid is the core CLI tool of Factory, and its user experience is similar to Claude Code or Codex CLI.
Features such as Plan mode, custom commands, and custom agents are fully supported, so I think you can switch from CC (Claude Code) without any discomfort.

Furthermore, it integrates with the web app via Factory Bridge, featuring capabilities such as viewing CLI session history on the web or requesting tasks from the web to edit code locally.
Moreover, even without using your own PC, you can have Droid perform tasks from the web using remote workspaces.

A VS Code extension is also available, allowing you to operate Droid within VS Code.

🤝 Tool Integration

While MCP is naturally supported, representative tools can be easily integrated via the web to execute powerful workflows.
This allows you to assign tasks to Droid for autonomous development while managing tasks in Linear or Jira without the need for complex settings, and you can also
request work from Slack.

👍 Points I Like

I'll share some points I've particularly liked after using it for about a month.

High Freedom in Model Selection

OpenAI GPT-5, Claude Sonnet, and Opus are all available.
It immediately supported Sonnet 4.5, so you no longer have to worry about whether Claude Code or Codex is superior. You can use both within the same price plan!
Also, although limited to the CLI, a cost-efficient model called Droid Core (GLM 4.6) has been added, expanding the range of choices.

It is also possible to leverage the strengths of each model, such as having GPT-5-Thinking brainstorm specifications and tasks, while having Claude Haiku or GLM 4.6 handle the coding.

Reducing Prompt Input and Improving Output Accuracy

First, I ask Droid to conduct research and define specifications, and based on those results, I create tasks in Linear following a template.
Then, I assign Factory to the issue in Linear, or I request work by passing the Linear issue as context in a prompt.
By doing this, I can minimize inconsistencies in output caused by vague instructions while streamlining the task request process.
As a bonus, I can do "laid-back coding" from my smartphone using Linear or Slack without even opening my PC!

It's a very good experience.

⚠️ Points to Note

Since the service is still new, there are a few things to keep in mind.
Here are the points I've particularly noticed while using it:

• There is limited secondary information available in Japanese.
• Integrations with Slack or Linear can sometimes be unstable.
• Due to active development, breaking changes may occur.

However, its appeal far outweighs these points, which is why I continue to use it!

🎉 Conclusion

Personally, I feel that Factory has great potential as a platform for accelerating AI-driven development.
I plan to continue using it and writing articles about it, so I encourage everyone to give it a try!