ZenCoder has updated its artificial intelligence (AI) platform for writing and testing code to provide integration with third-party DevOps tools such as JIRA, GitHub, GitLab and Sentry in addition to tighter coupling with VS Code and JetBrains integrated development environments.
At the same time, the company has added a “Coffee Mode” capability that now allows Zebcoder AI agents to perform tasks in the background.
Finally, Zencoder reports that with this update it has, based on the SWE-Bench-Multimodal benchmark, improved overall performance by a factor of two.
ZenCoder CEO Andrew Filev said the overall goal is to accelerate application development by integrating the company’s namesake AI platform more deeply into DevOps workflows.
The Zencoder platform, in addition to generating code, is also able to repair code, create tests and optimize code in real time using built-in static analysis capabilities. It makes use of multiple proprietary and open-source large language models (LLMs) to train agents to handle each of those tasks.
Rival approaches are, as a result, more prone to hallucinate in ways that generate flawed code, noted Filev. Zencoder solves that issue by providing development teams with a set of AI agents that can apply static analysis to code as it is being developed. For example, a Repo Grokking agent analyzes the entire code repository to provide the appropriate level of context required to generate code before an Agentic Repair agent creates a pipeline that automatically analyzes, fixes and refines generated code.
Most other AI tools don’t provide that level of context, resulting in code that actually won’t run in its intended production environment, noted Filev. Application developers are also unable to debug that code simply, because they don’t have any insights into how it was created by the AI tool, he added.
It’s not clear how many organizations are relying on AI to help build applications faster. A recent Futurum Research survey finds 41% of respondents expect generative AI tools and platforms will be used to generate, review and test code, while 39% plan to make use of AI models based on machine learning algorithms.
There is at this point no shortage of AI tools and platforms for writing and testing code. The challenge now is deciding which one best suits the needs of the organization versus any given individual developer. Unfortunately, many organizations still lack a coherent strategy so, as a result, many individual developers are experimenting with different tools and approaches without any overall organizational cohesion.
It’s not so much a question of whether or not AI will be relied on to build and deploy applications so much as it is the degree. Each application development team should, at the very least, now be experimenting with these tools. Actual usage of them to write code that will actually run in a production environment, however, will still require humans to know how that code has been constructed.
There is, of course, always pressure to build and deploy software faster. The issue is that it’s one thing to make a mistake when prototyping an application. It’s quite another to discover that mistake after an application has already been deployed in a production environment.
