Introduction
Agentic development is reshaping how we build software. Instead of writing every line of code yourself, you collaborate with AI agents that can plan, execute, and refine tasks autonomously. Inspired by the Spotify × Anthropic Live session, this guide walks you through adopting an agentic workflow. Whether you’re a solo developer or on a team, these steps will help you integrate AI agents without losing control over quality and direction.
What You Need
- AI Agent Platform – Access to a capable agent, such as Anthropic’s Claude (via API or chat interface) or a similar tool that supports code generation, debugging, and task decomposition.
- Development Environment – A local or cloud-based IDE (e.g., VS Code with an agent plugin) with version control (Git) and CI/CD pipeline.
- Project Repository – A codebase where you can safely experiment. Start with a small, non-critical project.
- Clear Task Definitions – A list of well-scoped features or bugs that can be delegated to an agent.
- Review Process – Code review guidelines and tools (e.g., GitHub pull requests) to maintain quality.
- Ethical Considerations – Understand data privacy, bias, and security implications of using AI agents.
Step-by-Step Guide
Step 1: Define Agentic Objectives
Before running an agent, clarify what you want it to accomplish. Break down a feature into sub-tasks: research, design, implement, test, and document. For example, instead of saying “build a login system,” specify: “create a secure JWT-based authentication with email verification.” This narrows the agent’s scope and improves output quality.
Step 2: Set Up the Agent Environment
Configure your IDE or command line to connect to the agent API. In Spotify’s case, they used Anthropic’s API to power agents that interact with their codebase. Install necessary packages (e.g., anthropic Python SDK). Create a configuration file with your API key and define system prompts that set the agent’s role (e.g., “You are a senior Python developer working on a microservice”).
Step 3: Delegate a Task with Clear Instructions
Write a detailed prompt for the agent. Include the task description, relevant files, coding style, and constraints. For instance: “Implement a function that validates user input (email regex) and returns a structured JSON. Use pydantic models. The function should be in utils/validators.py.” The more context you provide, the fewer iterations the agent needs.
Step 4: Review and Refine Output
After the agent produces code, treat it as a draft. Run automated tests, check for security flaws, and review code manually. If the output is incorrect, provide feedback: “The regex allows spaces – please fix it.” Use a conversational loop to refine until the solution passes all criteria. Never merge without human review – the agent is an assistant, not a replacement.
Step 5: Integrate into CI/CD Pipeline
To scale agentic development, embed agent interactions into your continuous integration flow. Spotify used agents to automatically generate unit tests and fix linting errors. Set up a GitHub Action that triggers an agent to create a PR with suggested improvements. Ensure the human developer must approve the PR before merging.
Step 6: Monitor Performance and Adapt
Track metrics like success rate, time saved, and code quality. If an agent repeatedly fails on a specific task, adjust the prompt or break the task further. Over time, build a library of successful prompts and agent workflows that your team can reuse.
Tips for Success
- Start small – Delegate one simple bug fix or test case before moving to complex features.
- Maintain a safety net – Use version control and sandbox environments to experiment without risk.
- Document agent behavior – Keep notes on what prompts work best, so your team can collaborate with the agent consistently.
- Combine agents with human intuition – Use agents for grunt work (boilerplate, tests, refactoring) and focus your energy on architecture and business logic.
- Stay ethical – Review any data the agent accesses or generates, especially if it involves user information. Ensure compliance with your organization’s policies.
- Iterate on the agent itself – As the technology evolves, update your agent configuration to leverage new capabilities.
For a deeper dive, revisit Step 1 to align your goals with agentic development.