Tell AI to Fix a Bug in Slack and It Handles Everything Up to Submitting the Code — Open SWE Released
LangChain has open-sourced the design patterns behind AI coding bots used internally at Stripe, Ramp, and Coinbase. Open SWE lets you request code fixes from AI in Slack and automatically handles everything from editing code to creating pull requests — and it surpassed 5,700 GitHub stars within a day of its release.
Send a message in Slack saying "@openswe fix this bug", and the AI analyzes the code, makes the fix, and even submits a pull request (PR, a formal code change request) on GitHub — all on its own. This isn't science fiction. Major Silicon Valley companies like Stripe, Ramp, and Coinbase have already been running AI coding bots like this internally.
On March 17, 2026, LangChain, a company specializing in AI development tools, open-sourced this secret weapon for anyone to use. It's called Open SWE. Within just one day of its release, it surpassed 5,700 GitHub stars, drawing massive attention from the developer community.
Top Silicon Valley Companies Were Already Using AI Coding Bots
The reason Open SWE is getting so much attention is that it's not just another coding tool. This project is the result of analyzing real-world practices from three of the world's top tech companies and distilling their patterns into a single framework.
Stripe's Minions — The leading fintech company gives its AI access to around 500 tools and lets developers request code fixes through Slack. Their key principle: "It's not about piling on tools — it's about curating the right ones."
Ramp's Inspect — The financial startup Ramp automated a verification step where the AI visually confirms the results after making code changes.
Coinbase's Cloudbot — The cryptocurrency exchange Coinbase runs a system where multiple AI agents cross-review each other's code before automatically merging it.
Open SWE extracted the 7 common design principles found across all three cases and turned them into a framework that any organization can use to build their own AI coding bot.
How a Single Slack Message Gets Your Code Fixed
Open SWE's workflow is intuitive.
Step 1: Make a request to the AI from Slack, Linear (a project management tool), or GitHub.
Example: @openswe repo:our-company/backend Fix the bug where no error message appears when login fails
Step 2: The AI analyzes and fixes the code inside an isolated cloud environment (a sandbox — a safe, contained workspace). Your company's actual servers are never affected.
Step 3: Once the fix is complete, it automatically creates a pull request (PR) on GitHub. A human reviews and approves it before it goes live.
You can even send additional instructions via Slack while the AI is working. For example, if you say "Oh, and add test code too," the AI picks it up in the next step. It also supports parallel execution, meaning it can handle multiple requests at the same time.
How Is This Different from Existing AI Coding Tools?
AI coding tools like Cursor or Claude Code are designed for individual developers to use on their own machines. Open SWE solves an entirely different problem.
Personal tools (Cursor, Claude Code) → An AI coding assistant you use solo on your own computer
Team tools (Open SWE) → An AI developer bot your entire team uses together in your company's Slack
Here are the key differences:
Only 15 curated tools: Stripe gave its AI 500 tools, but research showed that "the quality of tool selection matters more than the number of tools." Open SWE applies this lesson by including only 15 essential tools.
Works with any AI model: You can choose from Claude (Anthropic), GPT-4o (OpenAI), or Gemini 2.5 Pro (Google). You can even mix models — using a fast model for simple questions and a smarter model for complex code changes.
Flexible execution environments: It supports various cloud sandboxes (isolated work environments) like Modal, Daytona, Runloop, and LangSmith. You can also connect your own servers.
A 7-Layer Architecture — Designed for Safety
Having AI directly modify company code can be risky. Open SWE addresses this with a 7-layer safety architecture.
The "safety net" middleware is particularly noteworthy. Even if the AI finishes fixing the code but forgets to create a PR, the system automatically detects this and submits one on its behalf. It's a design that compensates for AI mistakes at the code level.
How to Adopt It at Your Company
Open SWE is released under the MIT license (free for commercial use). Setup requires Python 3.11 or higher, a GitHub App configuration, and a LangSmith account.
# 1. Clone the repository
git clone https://github.com/langchain-ai/open-swe.git
cd open-swe
# 2. Set up the environment
uv venv && source .venv/bin/activate
uv sync --all-extras
# 3. Start the server
uv run langgraph dev --no-browserOnce you connect a Slack bot, Linear webhook, and GitHub App, your team can start using it right away. For production environments, you can deploy via LangGraph Cloud, or host it on your own servers.
By placing an AGENTS.md file in each project folder, the AI automatically learns that project's coding rules and style. This enables project-specific customization without any additional configuration.
The Era of AI Coding Bots Joining Your Team
The arrival of Open SWE signals where AI coding tools are headed. The AI coding tools that individuals used in their terminals are now becoming Slack teammates for entire organizations.
LangChain built Open SWE on top of Deep Agents (a personal AI coding agent) that they previously released. If Deep Agents was "an AI developer for me alone," Open SWE is "an AI developer for our whole team."
That said, installation and configuration currently require technical knowledge. It's best suited for organizations with a development team that can handle GitHub App creation, webhook setup, and API key management. For individual developers, existing tools like Claude Code or Cursor remain the more convenient option.
Related Content — Get Started with AI Using EasyClCode | Free Learning Guide | More AI News
Sources
Stay updated on AI news
Simple explanations of the latest AI developments