[AIEWF Preview] Containing Agent Chaos — Solomon Hykes

Alessio, partner and CTO at Decibel, engages in a discussion with Mikolas Wicks, founder of Small AI, and Solomon Hykes, creator of Docker. Solomon introduces Dagger, a workflow engine and automation tool aimed at improving software delivery efficiency by transforming semi-automated workflows into robust, modular workflows that operate in containers. Dagger is open-source and supported by a community of platform engineers focused on enhancing developer productivity.

The conversation emphasizes Dagger's focus on post-development processes, addressing inefficiencies in current delivery methods. Solomon notes that recent market shifts involving AI agents have integrated Dagger into the development loop, despite its initial focus not being on AI. The platform is particularly useful for building CI/CD pipelines, creating reliable and portable environments for workflows. The rise of coding agents is changing developer roles, shifting them from solely writing code to managing these agents.

There is a growing interest in integrating large language models (LLMs) into workflows, as developers evolve into platform engineers capable of enabling AI agents to enhance productivity. The discussion also highlights the current state of coding environments, where experimentation is common but lacks clean isolation. The goal is to establish perfectly isolated, reusable, and portable environments to avoid dependency on specific applications, models, or cloud providers.

Solomon reflects on historical technology waves, like Docker, which faced fragmentation due to cobbled-together platforms. He emphasizes the need for open-source standards to improve coding agent environments, as many currently operate under limitations such as single VM access and no internet connectivity. He advocates for a rethinking of tooling to address these challenges, stressing the importance of isolation in development environments.

Despite Docker's initial success, the ecosystem has seen stagnation and fragmentation in development tools, with most work still occurring on local machines rather than in containers. Solomon outlines criteria for an ideal developer experience, including well-isolated environments for parallel agent work, portability across models and cloud providers, full observability of environments, and strong multiplayer elements for agent-human interaction.

Concerns are raised about the trend towards highly integrated, vertical end-to-end monoliths in software development, particularly the lack of standards in environments for coding agents. Limitations of Docker are discussed, such as agents' inability to make changes within containers and the absence of an AI-native interface. Solomon calls for a redesign of the development process to improve user experience, leveraging existing standards like container technology, Git, and APIs.

The conversation also addresses the challenges of running coding agents, including ensuring repeatability and traceability of artifacts, and the emergence of ephemeral apps that complicate application setup and teardown. Users often avoid containers due to their complexity, leading to makeshift solutions that fail to address issues like file system isolation and the need for multiple agent instances.

Local execution is crucial for developers, yet current CI/CD vendors lack effective solutions, complicating the standardization of development environments for coding agents. There is a call for major cloud providers to adopt tools like Dagger, which aims to create modular systems that integrate with existing infrastructures. The discussion concludes with a focus on the importance of community and ecosystem support for adoption, noting that while larger companies have advantages, startups can gain momentum without excessive gatekeeping.