theCUBEd Awards | AI Chat by Tabnine (Top AI Innovation for Developers)

Clips
News
More from theCUBEd Awards

Ameya Deshmukh

Head of Enterprise & Marketing Programs

Tabnine

Tailored AI for developers: How Tabnine optimizes coding with context

Tabnine Ltd. wants to offer the best AI agents in the industry, neatly wrapped up in an advanced context engine.Tabnine originated the artificial intelligence code assistant category with its first model in 2018. Since then, the company has grown into an AI software development platform designed to meet the needs of enterprise engineering teams.“What we uniquely offer [customers] at Tabnine is a platform that’s entirely tailored to you,” said Ameya Deshmukh (pictured), head of marketing programs at Tabnine. “They tell us what models that they’d like to use and how they’d like to deploy them

Tailored AI for developers: How Tabnine optimizes coding with context

Tabnine Ltd. wants to offer the best AI agents in the industry, neatly wrapped up in an advanced context engine.Tabnine originated the artificial intelligence code assistant category with its first model in 2018. Since then, the company has grown into an AI software development platform designed to meet the needs of enterprise engineering teams.“What we uniquely offer [customers] at Tabnine is a platform that’s entirely tailored to you,” said Ameya Deshmukh (pictured), head of marketing programs at Tabnine. “They tell us what models that they’d like to use and how they’d like to deploy them

play_circle_outline Tabnine's origin and innovation in AI code assistance

play_circle_outline Balancing privacy, performance, and security in AI software development

play_circle_outline Benefits of AI agents in every step of software development lifecycle

play_circle_outline Secret sauce: personalized, private, and compliant AI platform for enterprise teams

Info
Transcript

AI Chat by Tabnine (Top AI Innovation for Developers)

Ameya Deshmukh

Head of Enterprise & Marketing Programs Tabnine

Ameya Deshmukh, CEO of Tabnine, joins theCUBE Research team to discuss Tabnine's achievements and their unique contributions to Artificial Intelligence (AI) in software development. As the originator of the AI code assistant category, Tabnine is recognized for its innovation in AI chat for developers. The interview is conducted by analysts from theCUBE, focusing on Tabnine's evolution and how it supports enterprise engineering teams across diverse, regulated industries.

Deshmukh elaborates on Tabnine's accomplishments, including the flexibility of thei... Read more

explore Keep Exploring

What advancements has Tabnine made in the field of AI code assistance and how are they tailoring their product for enterprise engineering teams? add

What are some common limitations and trade-offs associated with AI coding systems on the market, and how does the Tabnine AI software development platform address these issues? add

What steps do Tabnine's AI agents assist customers with in the software development lifecycle, and what productivity gains are typically seen by customers? add

What sets Tabnine apart in the market and what do they offer to mature enterprise engineering teams? add

bolt Powered by CUBE AI

AI Chat by Tabnine (Top AI Innovation for Developers)

search

>> Kristen Nicole Martin here with theCUBE. We have another great guest, another award winner for the Top AI Innovations Awards for developers. Here with us today is Ameya Deshmukh. We're here to talk a little bit about the category that you guys won, and you also have a very unique story in that you originated one of the AI chat categories, so would love to hear about the company and what that means as far as what you guys have originated here and a little bit about what you guys do.

>> Yeah, absolutely. So Tabnine was actually the originator and creator of the entire AI code assistant category, launching our first code completions model back in 2018. Since then, we've led the market in delivering innovative capabilities that meet the needs of enterprise engineering teams. We followed that innovation up with announcing AI chat back in 2023. And in 2024 we saw massive upgrades to the product capabilities, including the initial release of our enterprise create advanced context engine, as well as AI chat agents for every step of the SDLC. And we're quickly involving into a AI software development platform that's tailored for the needs of a true enterprise engineering teams, and we're delivering the industry's most contextually aware AI agents for software development.

>> That's all really great stuff. And so tell me a little bit more about how you guys reached this point and what were some of the key points of development for you guys to reach the Innovations Awards here?

>> Absolutely. So originally when Tabnine started, we saw massive adoption among the developer audience. Since then, we are in use by over 1 million active users per month, and we are second in the market to only get up Copilot in terms of enterprise product adoption. Primarily our customers are in highly regulated and privacy conscious industries. So we're currently working with many of the largest players in aerospace and defense financial services and banking, pharmaceutical manufacturing and healthcare as well. Now, what kind of led to the innovation award victory for us was three things. So one thing we noticed was that all of the AI coding system players on the market force enterprise teams to trade off between privacy, protection and performance. So with the Tabnine AI software development platform, you no longer have to make that trade off. Second, we noticed that the majority of the LLMs in generic AI code assistance are merely thin application layers around an LLM. And the result from that has been that enterprise usage of these tools has actually scaled tech debt, not code quality. So what we've been really focused on as a platform is contextualizing the response and tailoring it to meet that organization's specific code quality security, architecture and maintainability standards so that when you're using Tabnine and using our AI, you're scaling code and you're scaling high quality code that meets your specific standards and is compliant with your organization's policies. The third thing we noticed was that especially in highly regulated industries, there is a need for flexibility in the platform. So there's two important factors that have occurred here. One is the emergence of open source large language models that are highly performant in 2025 has resulted in many of our enterprise customers starting fine-tuning and model customization efforts internally where they're developing and developing and creating their own models. So with Tabnine, they now have the architectural freedom to input those models into our AI development platform and make use of our context engine and AI agents. Secondly, it's been well established in academic research and other studies that LLM models on their own can generate code that is IP liable inducing at a rate of 0.88% to 2.01%, as well as generic AI code assistance have been shown to generate code that has security vulnerabilities in it. So we've delivered on innovations in this space as well that protect our customers through provenance and attribution and censorship, protecting them from any risk of IP liability code injecting into their code bases. And then secondly, our code review agent has enabled our customers to use AI to not only generate code but validate, and reconfigure that code to meet their specific security and performance standards.

>> So you had a lot of great stats there and use cases particularly for the areas that need that level of oversight and regulation. And with last year being so full of excitement and hype for agentic AI in particular when it comes to the LLMs and really ensuring quality there, that's certainly an area of innovation and one of the reasons why it was so important and special that you guys are getting this award and being recognized for that. So what are some additional use cases that you guys could maybe talk about in that regard? Because like you said, making sure the quality is there, making sure that you're able to fit in with these other industries that need that regulation, that's pretty crucial.

>> Yeah, so today, Tabnine's customers are using our AI agents across every single step of the SDLC and typically, the productivity gains that they're seeing a range from 30% to 50% increase in productivity across the full set of tasks that constitute the SDLC. So we've built a set of AI agents for every step. So whether your developers are onboarding onto a new project, looking to have code explained to them, or trying to generate code, or trying to fix code to resolve bugs and errors, or trying to get AI help to review code as well as generate first implementations for jury issues and tickets. And as well as test and update their testing frameworks, Tabnine's AI agents are helping them at every step of the way to deliver code that meets their patterns and approach and is in line with their company's standards of policies.

>> So tell me about, you've given so many great examples, and we probably already touched on some of the key turning points, but specifically when it comes to areas that you feel are truly exemplary of the innovation components here, I'm sure there's one or two moments over the past few months as a startup in particular, what are some of your highlights?

>> Yeah, over the past few months, we've made significant upgrades to our advanced context engine. So now we support over 15 different data points for local IDE context. Additionally, we can connect into any Git-based repository and we offer unlimited repository connections to our customers and that supports GitLab, GitHub and Bitbucket repos of any flavor and variety. Additionally, we also support connection in the non-code sources for context. So you can now connect Jira data center and cloud into Tabnine. Have Tabnine read your Jira issues and execute builds for review based off that. It's coming soon in our roadmap is the ability to connect into Confluence for even more non-code context. Additionally, a key innovation we delivered was our customizable code review rule sets. So our code review agent comes pre-built with thousands of rules from industry best practices. Additionally, our customers can source rules from their internal human experts, and the code review agent can be pointed at their own pull requests as well as code bases in order to ingest and identify rules for it to flag and fix. Now the fourth innovation that we've made to the context engine has been our support for model fine- tuning. So many of our customers, for example, in semiconductor manufacturing, have hardware engineering teams that are writing code and system Verilog and VHDL. Now the issue is that none of the LLMs available on the market today can actually perform to the level that they need to support those languages. So what Tabnine has offered and built for them has been a fine-tuned small language model for code completions that's uniquely tailored to them, supports their specific code base and their particular usage of that language. So we've been able to create really high precision output from AI for working on mission-critical applications for these customers. And then of course, last week actually, we launched our context scoping capability, which lets individual developers have fine precision control over the sources of context that they want Tabnine to ingest and consider for every task at hand.

>> Well, nice segue into your recent news announcement there. Let's hear a little bit more about that because I'm sure it's very big important announcement for you guys.

>> So regardless of whatever agent that you're using and whatever step of the SDLC you're in, the challenge that presents itself for developers when using AI is managing all the various sources of context that this system can connect to. So with our context scoping feature, you can select images as context, you can select and layer on local workspace context from files, imported packages and libraries. You can even at-reference specific objects and pieces of your code. And then of course you can layer on the appropriate repository that you've connected to that you want the AI to consider. And really what we're focused on developing is a tailored experience for each individual developer that lets them use AI in the way that they see best fit. And the context scoping functionality really gives that fine-grained control over retrieval-augmented generation down to the individual developer's hands on a task level, which is absolutely fantastic for increasing the quality of AI output.

>> When it comes to all the amazing things that have happened with AI specific to enterprise and real world business use cases, the developer community is a very unique component here. And with all of the things that you just described in your announcement and really what you guys are looking to do overall, what are some of the more exciting things that you're hearing from the developer community about the innovations that you're working on?

>> So among our users, the features and functionality that they find most exciting have been something actually that I haven't spoken about yet, which is our custom chat behavior and custom commands capability, which we released in January. So what this has really unlocked for them is the ability to create project-specific workflows using our AI system that they can then share with their team members. So one thing we've seen our customers do is, for example, create a custom command to reference a specific DoCmd file that has guidance on exactly how documentation needs to be written for the specific project. And then they then instruct the documentation agent to reference that style guide and then create documentation that matches that for that specific project across multiple files. And this has been a massive time saving for developers. If you ever talked to a dev, they love shipping code and building new features, they hate writing documentation. Additionally, the next agent that they've been most excited by has been our test case agent. And what this has really allowed them to do is through accessing the beaker icon inside their IDE, they're able to direct the agent at a existing file and then create comprehensive testing plans that are shown to them in plain natural language, and then have the agent actually go off and build and implement those tests. And what we're seeing them do is run that agent over and over again on their testing files to continually update their test coverage. So test coverage is always a challenge to maintain. And with this testing agent, our customers are really able to do that at a scale and with the ease of use that has never been given to them before. And then of course our Jira decode agent has been a big hit in success as well. This is really remarkable because it's no longer requiring developers to constantly contact switch in and out of their IDE to go read Jira tickets and then go back to the implementation. It puts their tickets and issues that have been assigned to them right into their IDE chat. All they have to do is at reference and select the specific issue they want to work on. Tabnine then goes and reads the context in the Jira issue, generates a first case implementation of that issue, tells the developer whether it meets all of the acceptance criteria, and then all they have to do is click our apply button and Tabnine automatically puts the code where it's supposed to be and shows them the diff in the file so that they can review it prior to accepting it. So it's a fully human in the loop experience that's really streamlining execution on Jira tickets, which is critical for developers to be able to meet their sprint deadlines and satisfy their engineering managers.

>> And you've explained so many great things about Tabnine today, we are asking all of our awardees, there's secret sauce that every successful company has when it comes to innovation. And maybe you've already given us a few of those ingredients, but what else can you kind of tell us about what the secret sauce is at Tabnine?

>> Yeah, the secret sauce at Tabnine is really our focus on meeting the needs of mature enterprise engineering teams. And what we uniquely offer them at Tabnine is a platform that's entirely tailored to you. So they tell us what models that they'd like to use and how they'd like to deploy them. They tell us how they'd like to deploy our AI platform, whether that's SaaS, VPC, on-prem or air gapped, or any combination of the above within a single organization. They tell us what repositories they'd like to connect to, and they even have total customization over the behavior of all of our agents as well. So in net, we're focused on delivering a highly personalized experience that is totally private and totally protected, that lets them adopt AI while remaining compliant with all their organizational guidelines. And really, I'd say the secret sauce that we have is our advanced context engine. So Tabnine, for example, you turn on the first level of our context engine and we see a 82% lift in chat consumption rates and a 36% lift in code completion consumption rates over baseline LLM performance alone, which that's all the developers are really looking for, right? Like Query and AI. They want to get an answer that they can actually use. And with Tabnine they get that.

>> Congratulations again on winning CUBEd Awards. And before I let you go, let me get one final thought from you on how you envision the future of agentic AI?

>> Absolutely. So the way we see the future of agentic AI playing out is through developing human-in-the-loop AI agents. So it's critical that developers and broadly organizations have full visibility into the behavior and thinking of their AI agents as well as the ability to define the scope in which that they can operate. And all of our AI agents we built with this human-in-the-loop design principle in mind because certainly you can use any generic agentic AI system to go from a blank IDE screen to a first pass of an application, but our customers are working on applications with millions of lines of code in them already. And for them, it's critical to be informed and be fully in control of any agentic AI. And that's really what we believe the future is for enterprise agentic AI usage.

>> Well, thank you so much for sharing your story. Thank you for being part of theCUBE community, and we're looking forward to hearing so much more from your startup.

>> Thank you.

>> Thank you all for watching. This is Kristen Nicole Martin with theCUBE, and stay tuned for more from our award winners.