Tabnine today revealed that it is giving DevOps teams the ability to switch between multiple large language models (LLMs) when they use its generative artificial intelligence (AI) platform to write code.

Application development teams can continue to use the LLM that Tabnine originally developed for its Tabnine Chat tool for generating code using a natural language interface. Alternatively, they can invoke either an instance of open source Mistral LLM jointly developed with Mistral AI or version 3.5 or 4.0 of the ChatGPT LLM (created by Open AI).

Tabnine president and chief marketing officer Peter Guagenti said the company will continue to add support for other LLMs, given the pace at which multiple LLMs advancements continues to accelerate.

As part of that effort, Tabnine will surface insights into LLM behaviors and their performance, privacy and protection behaviors, so development teams can identify which LLM is best for a specific use case, Guagenti added

Tabnine created an AI agent optimized for software development that optimizes prompts, provides contextual awareness of local and global code bases, and fine tunes LLM output.

There is, of course, no shortage of platforms for generating code, but the LLMs that Tabnine created are trained using a narrower base of code to reduce errors and hallucinations. However, development teams sometimes prefer to use LLMs that can support more parameters. Of course, the more parameters an LLM has, the more expensive it becomes, as the IT infrastructure resource requirement increases.

Regardless of the LLM type, the secret sauce is how they are invoked rather than in the LLMs that are rapidly becoming commodity services, noted Guagenti. Last month, for example, Tabnine made it possible for application development teams to better personalize output by exposing an LLM to both Tabnine’s code base and development environments.

It’s now only a matter of time before generative AI is applied across the entire software development life cycle (SDLC). In the short term, though, the pace at which code is being written should accelerate. The quality of that code will vary depending on the ways those LLMs were trained. A general-purpose LLM such as ChatGPT typically was trained using code examples of varying quality collected from across the Web. As such, that LLM is more likely to convincingly present code that doesn’t actually work, might run inefficiently, or contain vulnerabilities that could be exploited by cybercriminals.

The expectation is the output from LLMs that were been trained using vetted code will provide more reliable results. The challenge and the opportunity then becomes determining how best to apply AI to DevOps workflows before those teams are overwhelmed by the volume of code being generated by machines.

It may still be a while before AI is pervasively applied across the entire SDLC but the genie is out of the proverbial bottle. Eventually the overall code quality that makes it into production environments will improve steadily. In the meantime, DevOps teams should put guardrails in place to make sure the code AI is generating in the short term actually meets their requirements.