Docker has made its new Bake feature generally available, and released Docker Desktop 4.38 including a preview of AI Agent, codenamed Gordon.

Docker Bake simplifies building docker images by using a build configuration file in place of multiple arguments to the docker build command. Bake configuration files can be in HCL (Hashicorp Configuration Language), YAML, or JSON format, though HCL is recommended. The key Bake document, the configuration file reference, remarks that “aside from syntactic differences, HCL lets you use features that the JSON and YAML formats don’t support.”

Bake build targets support a range of attributes including dockerfile, platforms, output destinations, and attestations, which allow inspection of the SBOM (Software Bill of Materials) and the provenance of a container image. Bake groups allow multiple targets to be invoked at once. Bake also supports variables, and functions including general-purpose and user-defined functions. 

Although Bake does not add new features to Docker, it makes complex builds easier to maintain and to share with a team. In many cases Bake will be integrated into continuous integration tools such as GitHub Actions, which has a bake-action that can be included in workflows. According to one developer, Bake is “an unsung hero when it comes to building multiple images in parallel,” needed for multi-container applications such as those defined by Docker Compose.

Docker has included Bake in the just-released Docker Desktop 4.38, though Desktop is not required to use it. 

Desktop 4.38 has another new feature though, which is a preview of AI Agent, also known as Gordon. The preview is available to all signed-in users, but disabled by default, perhaps because the agent “may send relevant files, directories, or image data to the Gordon backend along with your query,” as the docs state. Data is encrypted in transit but Docker collects “anonymized data” from interactions with Gordon, though it is not used to train AI models. 

The purpose of the AI Agent is to be an expert assistant, with guidance on how to use Docker including optimization, looking up documentation, how to containerize an application, help on Docker in GitHub Actions, and suggested fixes when containers do not start.

The documentation adds cautions, including that the answers may be inaccurate and should be verified. There is, perhaps, a difficulty here in that a developer turning to the agent for expert assistance may not be best placed to verify the accuracy of the answer; but Docker is at least giving due warning.