Tabnine today added artificial intelligence (AI) agents to its namesake platform for generating code that DevOps teams can now use to automate workflows.
Company CTO Eran Yahav said Tabnine Agentic is based on the existing Context Engine embedded in the platform that is being extended to enable AI agents to understand and reason across repositories, tools, and policies to enable them to plan, execute and validate multi-step development tasks, including refactoring, debugging and creating documentation.
Previously, Tabnine had provided an AI capability that provided code suggestions. The company is now moving into the agentic AI era by adding a set of autonomous agents that can be assigned entire workflows, said Yahav.
The Tabnine Context engine is designed to adapt to new codebases and policies without retraining or redeployment by applying vector, graph, and now agentic retrieval techniques to a large language model (LLM). That approach makes it possible to adapt to changes in the development environment because the AI agents are not dependent solely on when an underlying LLM was last retrained, noted Yahav.
Because it does not package a specific LLM with its platform, Tabine is also to provide a straightforward pricing model based on a flat monthly fee for access and the actual usage of its AI agents, he said. Instead, DevOps teams can connect its Context Engine to their preferred LLM. If DevOps teams decide to use an LLM through Tabnine, billing is handled on a “pass-through” basis, subject only to a nominal handling fee. DevOps teams can also apply customizable quota limits by team or business unit to limit their total costs.
Finally, centralized controls for applying governance policies ensure permissions and usage follow policies defined within the Context Engine to provide auditing capabilities in a way that makes it possible to adhere to compliance mandates, he added.
As organizations rely more on AI tools and platforms to generate code, managing DevOps workflows is becoming more challenging and, arguably, more costly. Tabnine is making a case for an approach to using AI to write code that provides DevOps teams with an ability to more granularly control how agents are being used to better control costs.
DevOps teams that lack those controls will also soon find themselves drowning in a tsunami of technical debt as AI agents continue to generate massive volumes of code without any controls being applied, said Yahav.
Each DevOps team will need to decide for itself to what degree to rely on AI agents to generate code. As the amount of memory AI agents can use increases, so too does their ability to reason. However, AI agents tend to have a voracious appetite for data, so without the proper controls in place, the output being generated may include, for example, sensitive data that the agent should never have been allowed to discover.
Regardless of these and other concerns when using AI to generate code, there is no going back. The AI genie is not going back in the proverbial bottle, so the issue now is determining how best to govern what will soon be potentially hundreds of AI agents that will soon be strewn across the software development lifecycle (SDLC).
