Autonomous Agents in Developer Tooling: Key Insights and Best Practices

The landscape of developer tooling has seen rapid evolution over the past decade. Ten years ago, we were excited about autocomplete and intellisense, which suggested method names and performed programmatic refactors. Four years ago, copilots and tab complete took the stage, writing a few lines of code for us. Two years ago, generative chatbots entered the scene, assisting in development and generating entire files. Today, we're witnessing the rise of autonomous agents-tools that can take initial descriptions to final pull requests with minimal human intervention.

At Devin, we've been working on realizing this vision for the past two years. Interest in autonomous agents is surging, particularly with recent releases from companies like OpenAI (Codex) and Google (Jules). Additionally, local agents such as Cursor and Claude Code can be run in parallel workspaces to achieve similar effects. These agents are available in various forms, including web apps, mobile apps, and integrations within popular tools like Slack, GitHub, Linear, and Jira.

While a human paired with an AI assistant is more powerful than any AI alone, the ability of autonomous agents to handle tasks end-to-end introduces a new level of multitasking. This transformation turns every engineer into an engineering manager, capable of overseeing multiple projects simultaneously.

However, adapting to working effectively alongside these new AI colleagues can take time. We've observed that senior-to-staff level engineers tend to adopt and become proficient with these tools the fastest. Based on our experience and customer feedback, here are key insights and best practices for successfully integrating autonomous agents into your workflows:

1. Specify How You Want Things Done, Not Just What

Think of the agent as a junior coding partner whose decision-making can be unreliable. For simple tasks, direct descriptions work well, but for more complex tasks, clearly outline your preferred approach from the outset. Providing the agent with the overall architecture and logic upfront boosts its chances of success and reduces your time reviewing code.

Example:

• Instead of: "Add unit tests."
• Specify: "Add unit tests for the UserAuthentication module. Test edge cases such as empty input, invalid tokens, and maximum login attempts. Mock the database calls to ensure isolation."

2. Indicate Where to Start

Consider where you would start if handling the task yourself. Even if you don't know specific file or function names, mention the repository, relevant documentation, and key components involved. Clearly indicating these elements minimizes wasted effort and confusion.

Example:

• Instead of: "Add support for Google models."
• Specify: "Please add support for Google models to our code. You should look at the latest docs here and create a new implementation file in the model_groups directory."

3. Practice Defensive Prompting

Imagine giving the same prompt to a new intern. Where would confusion arise? Anticipate potential misunderstandings and provide clear, detailed instructions.

Example:

• Instead of: "Fix the bug in the API."
• Specify: "Fix the bug in the getUserDetails API method. The issue occurs when the user ID is not found in the database. Ensure that the API returns a 404 error with a clear message."

4. Review and Iterate

Autonomous agents are powerful, but they're not infallible. Always review the agent's work to ensure it meets your standards. If necessary, provide feedback and iterate on the task.

Example:

• After initial implementation: "The getUserDetails method looks good, but can you add a log statement when a 404 error is returned?"

5. Leverage Agent Capabilities

Autonomous agents excel at repetitive, well-defined tasks. Delegate these to the agent so you can focus on more complex and creative work.

Example:

• Instead of: "Write the entire UserAuthentication module."
• Specify: "Generate the basic structure for the UserAuthentication module with placeholder methods for login, logout, and token validation."

6. Foster a Collaborative Environment

Encourage your team to experiment with autonomous agents and share their experiences. Collaboration can lead to better practices and more efficient workflows.

Example:

• Team meeting: "Let's try using the autonomous agent for our next feature.