Coding Agents

Coding agents let you delegate implementation tasks to external AI coding tools directly from the Modem Agent. Instead of switching between Modem and your coding tool, you describe what needs to be built and the agent handles the handoff. You can invoke coding agents from the dashboard (full-screen or sidebar) or from Slack. Modem currently supports two coding agents: Cursor (via Cloud Agents) and Devin.

​

How It Works

1. You ask the Modem Agent to implement something (e.g., “fix the login timeout bug”).
2. Before delegating, the agent queries your Modem data for relevant context: bug reports, user feedback, error patterns, affected user counts.
3. The agent composes a task description enriched with that context and sends it to Cursor or Devin.
4. You approve the action (all coding agent tasks require confirmation before executing).
5. The coding agent works asynchronously. You can check on progress through the Modem Agent.

​

Prerequisites

• A Modem organization with the GitHub integration connected (so the coding agent knows which repositories to work on)
• An API key for the coding agent you want to use (Cursor, Devin, or both)

​

Setup

Each team member configures their own API key. Keys are stored encrypted and never shared with other members.

​

Cursor

Cursor’s Cloud Agents run on Cursor’s infrastructure. They can implement features, fix bugs, create pull requests, and investigate codebases.

​

Getting your Cursor API key

1. Open cursor.com/settings/keys
2. Generate a new API key
3. Copy the key and paste it into Modem’s Coding Integrations section

​

What Cursor can do

• Implement features and fix bugs across your connected GitHub repositories
• Create pull requests with the changes
• Investigate codebases: read source code, trace execution paths, audit implementations
• Target a specific repository and branch

​

Example prompts

​

Checking task status

After triggering a Cursor task, you can ask the agent to check on it: The agent will return the current status, a summary of work done, and a link to the pull request if one was created.

​

Devin

Devin is an autonomous AI software engineer that handles complex, multi-step implementation tasks. Devin sessions run asynchronously and you can send follow-up messages to provide additional context or redirect its focus.

​

Getting your Devin API key

1. Open your Devin dashboard settings
2. Navigate to the API keys section
3. Generate a new API key
4. Copy the key and paste it into Modem’s Coding Integrations section

​

What Devin can do

• Handle multi-step implementation tasks across multiple files and repositories
• Create pull requests with the changes
• Investigate codebases: map architecture, audit implementations, report findings
• Accept follow-up messages during an active session

​

Example prompts

​

Checking session status

The agent returns the session status, recent messages, and a link to the pull request if one exists. You can also view the session directly on the Devin dashboard.

​

Context Enrichment

When you ask the agent to delegate a coding task, it automatically considers whether your Modem data can inform the work. For example:

• “Fix the login timeout bug” — the agent queries topics and discussions about login timeouts first, then includes error reports, reproduction details, and user impact in the task description.
• “Address the top complaints from this week” — the agent finds the most-reported issues and provides the coding agent with specific details.

If Modem data reveals ambiguity (e.g., multiple distinct issues matching your description), the agent asks you to clarify before delegating. You can also skip this by providing full context yourself. The agent only enriches when it would add value.

​

Troubleshooting

​

Related