What do Enterprises Need?

If you lead an enterprise that has code, you are always thinking of:

• Security of your code intellectual property
• Productivity of your software developers
• Quality of your codebase

And like any company, you are thinking about your strengths and weaknesses against competitors, and how you can gain an edge to remain or become the leader in your space.

Generative AI for code, such as Codeium or Github Copilot, increases the productivity of software developers and produces high quality code (having learnt from billions of lines of code). Companies that do not integrate generative AI capabilities into their workflows will lose an edge to those that do, which is why some select tech giants are investing heavily in integrating generative AI into not just their products, but also their internal processes. However, building AI code tools is not a core competency for almost every company, small or large, and buying a solution will get you state-of-the-art tools immediately while saving millions on development and maintenance costs.

Given needs, the ideal generative AI coding solution for enterprises would have:

• Maximum code security: code should never leave your private cloud or on-prem servers. This way your code can never be seen by anyone else, shared with anyone else, or used to train models that are used by anyone else. Doing this should not be cost prohibitive, but if it is and a managed solution is required, you should still have guarantees that your data is not sold, shared, or trained on.
• Maximum quality of assistance: any code suggestions or other assistance should be the best possible for your codebase, which can only happen by customization on your existing codebase, such as fine-tuning the models.
• Maximum productivity boost: code autocomplete is a good start, but software engineers can be supercharged not just by speeding up when they already know what to type. Software developers spend a lot of time generating test cases, writing documentation, searching through existing codebases, debugging, creating and reviewing pull requests, and more. Each of these different “modes” that a software developer operates in can be accelerated with AI.

Github Copilot for Enterprises does not help

Github Copilot has introduced an “enterprise plan,” but Github Copilot for Enterprises fails to address any of these enterprise needs:

• No guarantees on code security. Security is the same as on the individual plan, with code snippets being sent to Github, Microsoft, and OpenAI servers and unclear code retention and usage policies. You should have security and privacy concerns.
• No customization for your codebase. You only get the generic models and no customization whatsoever. It’s best practice for developers to use your company’s frameworks and libraries, and not doing so will decrease legibility and introduce technical debt, performance issues, and security holes.
• Only autocomplete. Copilot will only perform autocomplete and this functionality has not measurably improved over the last year or so. In addition, Copilot also does not work on every IDE and platform, such as notebooks like Jupyter or Colab, which is key for making all of your developers more productive.

All Github Copilot for Enterprises does is provide a team administrator to purchase and manage seats of Github Copilot for their employees.

Codeium for Enterprises does help

On the other hand, Codeium for Enterprises actually addresses your enterprise’s needs:

• Deployed entirely on-prem or in your Virtual Private Cloud (VPC). The best way to guarantee security is to not allow your data to leave your company’s managed resources. We have also trained models in house, built all IDE integrations, and written all custom logic to cleanly integrate the user’s code with model inputs and outputs. By not relying on third party APIs, you can be confident that there is no potential for external security vulnerabilities to creep in. We recognize that every company has different data handling and management policies, as well as hardware setups, so we offer a wide range of methods to deploy Codeium for Enterprises in a self-hosted manner. The compute requirements are also light, so you will not be breaking the bank to provide AI acceleration at lightning fast speeds and state-of-the-art quality. If you do not want to deploy locally, we do offer a managed service with SOC2 compliance coming very soon.
• Fine-tuning models on your existing repository entirely within your self-hosted Codeium instance. You are the industry experts in what you do, and have more relevant, higher quality code in your private repository than can be found in public repositories. Therefore, a generic model, irrespective of deployment method, would not provide the same generation quality as fine-tuned models. Codeium’s base quality is comparable with Github Copilot, and so fine-tuning will definitely outperform anything in existence for your needs. Doing fine-tuning within your self-hosted instance means none of your data will leave your management and no-one will gain temporary or permanent access, including us. These models will automatically be deployed in your self-hosted Codeium instance - the model weights will never leave your instance either.
• Full suite of AI acceleration for coding. We refer to Codeium as a toolkit rather than a tool since software development is not a homogenous endeavor. Each different “mode” that a software developer operates in can be accelerated with AI, and Codeium is the only product that will address this entire range in a consistent, seamless manner. Using a multitude of products to address each of these modes will expose you to colliding interfaces, larger management overhead, and higher security vulnerability. Already we provide Autocomplete on Enterprise, and we have Unit Test Generation, Doc Generation, and much more actively being worked on. We also have thoughts on Search for Enterprises (already provided to individuals) as well as Chat, PR Review Automation, Debugging, and more.

Case Study

The following case is from an actual existing Codeium user that agreed to provide us with code snippet telemetry.

The data scientist spends the vast majority of time working with Python:

While they primarily work in a standalone IDE, a nontrivial amount of time is spent in Jupyter and Colab notebooks, platforms that are only supported by Codeium:

We can look at a month of Codeium usage, computing the number of accepted completions per day and the number of characters in those completions:

The usage is spiky, with periods of lulls with little coding and then large bursts of coding (with recurring drops on weekends). Overall, Codeium’s autocomplete functionality saved this user 61,949 characters in a month. Given an average typing speed of 200 characters per minute, this directly corresponds to 5.16 hours saved in the month. This is likely a vast understatement of the amount of time saved by using Autocomplete:

• Assumes that engineer knows exactly what they want to type, just needs to execute, and can execute perfectly
• Ignores the extra time searched on Google or StackOverflow for solutions, API documentation, or unfamiliar language syntax
• Doesn’t include any time saved by search and other capabilities
• Reflects capabilities of generic models, not ones fine-tuned on your own repository

Codeium for Enterprises is worth it, Copilot is not

If you want to actually have a competitive advantage with AI without compromising your security, then Codeium for Enterprises is the direction your company should go with, not Github Copilot for Enterprises. We run all of our pilots for free since we are confident that all of your enterprise’s needs will be satisfied and that the return on investment will be obvious. Do not hesitate to contact us - we would be happy to discuss further, including pricing.