The integration of AI into software development has the potential to significantly reduce the time and effort required for manual tasks like writing unit tests and performing code reviews. My early explorations have demonstrated how we can leverage off-the-shelf AI tools together with the emerging Dagger open source framework to create scalable, distributed workflows that use AI to automate and improve these processes. Implementing systems with a self-corrective loop should lead to higher code quality while reducing the testing burden on developers.
Unveiling AI-Driven Workflows with Dagger
In the talk "Agentic Workflow powered by Dagger," Kambui Nurse provides an insightful exploration of integrating AI into development workflows, emphasizing practical applications through live coding and real-world examples. This session, hosted in Atlanta, was not only an educational experience but also a demonstration of the collaborative potential and challenges of agent-based developer tooling, specifically through the lens of the CoverAI project.
Building the Foundation: CoverAI and Dagger
Kambui Nurse begins by discussing the foundational elements of agentic workflows with a focus on Dagger, a tool that orchestrates containerized environments. This sets the stage for CoverAI, an AI agent adept at generating homepage introductions and automating software test creation and validation. Nurse’s approach—described as "vibe coding"—invites a hands-on participation from the audience, showcasing how minimal guidance can be sufficient for modern generative AI.
He emphasizes the importance of clear documentation and internal workflows, highlighting contributions by Tyrese Dixon, who provided essential README and documentation for CoverAI. “Let’s see if it gets it right,” Nurse remarks, as he tests the AI’s adherence to the provided guidelines, underscoring the necessity for comprehensive documentation in agent-driven tasks.
Orchestrating with Dagger
Central to Nurse's presentation is the synergy between Dagger and CoverAI. Dagger, in its version 18.5, facilitates the execution of automated tests within containerized setups, enabling seamless integration without manual intervention. Nurse elaborates on dependencies like OpenRouter, which serves as a unified interface for large language models (LLMs), illustrating the modularity and interoperability crucial in modern developer workflows.
Extensibility and Human Oversight
A significant focus is CoverAI’s extensibility, which supports Jest and PyTest while allowing additional plugins through specific functions: acquiring code tests, generating coverage reports, and parsing test results. Nurse envisions a future where these tools are agent-coordinated yet remain user-extendable. He stresses the importance of human oversight, noting, “These agents [just] be doing stuff, we don’t know what they do sometimes… you’ve got to have a human in the loop somehow.” This transparency ensures that while AI can automate tasks, human intervention remains crucial for contextual understanding and decision-making.
Real-World Challenges and Debugging
Nurse candidly shares his experiences, acknowledging the challenges that come with agentic workflows. Despite achieving significant test coverage ("95% coverage" and "351 passing tests"), the automated generation can lead to unfamiliar codebases, complicating debugging processes. His narrative on collaborating with the generative agent Lova Bull serves as a learning point, illustrating the potential pitfalls and the necessity for pairing programming and iterative testing.
Future Directions and Reflections
Concluding his talk, Nurse champions a balanced approach combining agent-driven automation with practical engineering practices. He advocates for test-driven development and open extensibility while highlighting the irreplaceable value of human oversight. His insights reflect both the promise and the evolving nature of agentic development tools, offering a blueprint for developers seeking to enhance productivity while navigating the complexities of AI-driven environments.
About The Speaker
Kambui Nurse
Founder, AI Agency services
Kambui Nurse is a self-taught technologist with 20+ years of experience. As Lead Staff Engineer, Innovation at Marsh McLennan (MMC), he drives advancements in generative AI and is developing an Agentic workflow using Dagger, a cutting-edge container pipeline tool. His contributions to the Dagger community earned him the title of the first-ever Dagger Commander, recognizing his leadership and mentorship. While new to public speaking, Kambui regularly presents internally at MMC, educating teams on emerging technologies and AI-driven innovations.
THE WEEKLY DIGEST
Subscribe
Sign up to be notified when we post.
Subscribe