AI Agents

Friday Dev integrates multiple AI agents to handle different types of development tasks.

What Are AI Agents?

AI agents are autonomous systems that can:

• Read and understand your codebase
• Plan implementation approaches
• Write and modify code
• Create pull requests
• Respond to feedback

Available Agents

Agent	Model	Best For
Friday Agent	Claude	Complex tasks, full autonomy
Gemini	Google Gemini	Fast tasks, free tier
Claude	Anthropic Claude	Safe, detailed work
GPT-5 Codex	OpenAI	Advanced reasoning
GLM 4.7	Zhipu AI	Multilingual, coding
Qwen Code	Alibaba	Coding workflows
Cursor	Cursor AI	IDE integration
GitHub Copilot	GitHub	Code completion
Droid	Multiple	Flexible backend

Choosing an Agent

Friday Agent (Recommended)

Best for most tasks. Combines Claude's capabilities with custom tooling.

✅ Complex multi-file changes
✅ Full codebase understanding
✅ Autonomous decision-making
✅ PR creation

Gemini

Great for quick tasks and users on a budget (free tier available).

✅ Fast response times
✅ Free tier available
✅ Good for simple tasks
⚠️ May need more guidance

Claude

Best for tasks requiring careful, safe changes.

✅ Detailed explanations
✅ Conservative approach
✅ Good error handling
⚠️ Slower than others

GLM 4.7

Excellent for multilingual projects and specialized tasks.

✅ Strong multilingual support
✅ Code generation
✅ Chinese language support

Running Agents

From UI

1. Open a task
2. Click "Run Agent"
3. Select your agent
4. Configure options:
   • Autonomy Level - How much freedom the agent has
   • Branch - Where to commit changes
5. Click "Start"

From CLI

# Run Friday Agent on task
friday-dev run --task 123 --agent friday

# Run with specific autonomy
friday-dev run --task 123 --agent gemini --autonomy workspace-write

Autonomy Levels

Level	Description	Use Case
workspace-write	Can only modify workspace files	Safe, recommended
skip-permissions-unsafe	Full system access	Trusted environments

Agent Output

When an agent completes, you'll see:

Logs

Step-by-step output of what the agent did:

[1/5] Reading task description...
[2/5] Analyzing codebase...
[3/5] Planning implementation...
[4/5] Writing code...
[5/5] Creating PR...

Diffs

Code changes made:

+ import { auth } from './auth';
+
+ export function login(email, password) {
+   return auth.signIn(email, password);
+ }

Actions

List of all actions taken:

• Files created
• Files modified
• Commands run
• PRs created

Monitoring

Real-time Logs

Watch agent progress in real-time:

# Stream logs from CLI
friday-dev logs --task 123 --follow

Pause/Resume

Pause an agent mid-execution:

1. Click "Pause" button
2. Review current state
3. Click "Resume" to continue or "Cancel" to stop

Best Practices

Clear Task Descriptions

Agents work best with clear, detailed descriptions:

## Task
Add password strength validation to registration

## Requirements
- Minimum 8 characters
- At least one uppercase letter
- At least one number
- Show strength indicator

## Files to modify
- src/components/RegisterForm.tsx
- src/utils/validation.ts

Review Changes

Always review agent-generated code:

1. Check the diff for correctness
2. Verify no sensitive data is exposed
3. Test the changes locally
4. Approve or request modifications

Start Small

For new projects, start with small tasks to calibrate:

1. Simple bug fixes
2. Add documentation
3. Minor refactors

Then move to larger features.

Troubleshooting

Agent Stuck

If an agent appears stuck:

1. Check the logs for errors
2. Pause and review state
3. Try a different agent
4. Break task into smaller subtasks

Poor Output

If output quality is low:

1. Add more detail to task description
2. Include relevant context/files
3. Try a different agent
4. Use a higher-capability model

Rate Limits

If hitting rate limits:

1. Wait and retry
2. Use a different API key
3. Switch to an agent with higher limits
4. Self-host for unlimited usage

Next Steps

• Friday Agent - Deep dive into the flagship agent
• Task Management - Create better tasks
• Code Review - Review agent output