AI Agents in Software Engineering: A Q&A with Spotify and Anthropic

In a recent live session, Spotify and Anthropic explored how AI agents are reshaping software development. From automating mundane tasks to redefining the developer’s role, the conversation offered fresh insights. Below, we address key questions that emerged from the discussion.

What exactly is agentic development, and how does it differ from traditional automation?

Agentic development refers to using AI agents—autonomous programs that can perceive, decide, and act—to assist in software creation. Unlike traditional automation, which executes predefined scripts, agents can adapt to new contexts, learn from past actions, and make independent judgments. For example, an agent might not only compile code but also recognize a recurring bug pattern and suggest a fix without explicit human instruction. This shift moves developers from being micro-managers of tools to strategic guides overseeing a team of capable AI collaborators.

How are AI agents changing the daily workflow of a software engineer?

According to the session, AI agents take over repetitive, low-level tasks such as writing boilerplate code, running test suites, and formatting documentation. This frees engineers to focus on architecture, user experience, and creative problem-solving. Additionally, agents can offer real-time code reviews, flag potential security vulnerabilities, and even simulate edge cases. The result is a faster iteration cycle and a reduced cognitive load, allowing developers to spend more time on high-impact work.

What role does Anthropic’s Claude play in this Spotify initiative?

Anthropic’s Claude is used as the underlying language model to power agentic behaviors. Spotify integrated Claude into their internal development tools, enabling agents to understand natural language requests, generate code snippets, and maintain context across a project. Claude’s safety-focused design also ensures that agents remain aligned with developer intent, reducing the risk of unintended behaviors or security flaws.

Can you provide a concrete example of an AI agent in action at Spotify?

One example shared was an agent that automatically refactors legacy code. The developer would describe the desired outcome (e.g., “convert this service to use the new API pattern”), and the agent would scan the codebase, identify dependent modules, rewrite the relevant functions, and submit a pull request with a summary of changes. The agent also ran integration tests and flagged any breaking modifications. This process, which previously took a full day, was reduced to under an hour.

What are the main challenges in adopting agentic development?

Key challenges include trust, debugging, and integration. Developers must learn to trust the agent’s output, especially for critical production code. Debugging agent-driven changes can be complex because the decision-making path isn’t always transparent. Additionally, integrating agents into existing CI/CD pipelines requires careful planning to avoid conflicts. Both Spotify and Anthropic emphasized the need for gradual adoption, starting with low-risk tasks and building confidence over time.

How might the role of a software developer evolve with widespread agent use?

The panel predicted that developers will become more like “AI orchestrators” or “prompt engineers.” Instead of writing every line of code, they will define high-level goals, review agent outputs, and ensure quality and safety. Soft skills—like clear communication, ethical reasoning, and system design thinking—will become increasingly important. The developer’s identity shifts from being a pure builder to a curator of intelligent tools who knows when to intervene and when to let the agent run.

What advice did Spotify and Anthropic give for teams starting with agentic development?

They recommend starting small: pick a well-understood, repetitive task and let an agent handle it. Measure the time saved and error reduction. Then gradually expand the agent’s autonomy while maintaining human oversight. Also, invest in training for developers on how to write effective prompts and understand agent capabilities. Finally, prioritize transparency—build mechanisms to log and review every agent action. This builds trust and helps teams learn what works in their specific context.