January 2024 Summaries

Codeium has secured a $65 million Series B funding round led by Kleiner Perkins, with contributions from Greenoaks and General Catalyst, valuing the company at $500 million. Codeium develops AI tools designed to enhance the productivity of software developers by providing intelligent code suggestions integrated with developers' codebases. With over 300,000 users, including Fortune 500 companies, Codeium's AI supports more than 70 programming languages across 40 IDEs, emphasizing seamless integration, security, and customization to meet individual and organizational needs. The company's unique approach involves developing proprietary Large Language Models and managing the entire infrastructure, allowing them to offer their services at a lower cost than competitors while maintaining high performance and security. As they envision further increasing developer productivity, Codeium plans to expand its AI capabilities to include designing software systems and addressing security vulnerabilities, with aims to integrate more deeply into existing developer tools. The recent funding will support growth in sales and engineering teams, driving the company's mission to embed AI throughout the software development lifecycle.

Clearwater Analytics, a leading SaaS fintech company specializing in automated investment accounting and reporting, has embraced AI-assisted code generation to boost developer productivity. With nearly 2,000 employees and oversight of over $6 trillion in assets, Clearwater sought innovative solutions and found that Codeium provided the optimal blend of security, ease of deployment, and productivity analytics. The adoption of Codeium has been met with positive feedback, with developers appreciating the support, training, and rapid feature rollouts such as chat integration with IntelliJ and VS Code. Kelly Honeycutt, Software Development Division Manager, highlighted the proactive support and responsiveness of the Codeium team, which is committed to continuous development and enhancement of its functionalities for Clearwater and other enterprise clients in the software sector.

Codeium is committed to enhancing developer productivity by leveraging existing company knowledge through integration with Atlassian's suite of tools, including Bitbucket, Confluence, and Jira. This partnership allows businesses to adopt a secure and customized AI assistant that integrates seamlessly with their current workflows without needing to migrate their source code management. Codeium, which is available on the Atlassian marketplace, already supports various use cases, languages, and IDEs used by Atlassian customers. Future plans include even deeper integration with Atlassian tools to improve processes such as web PR reviews and ingestion of Confluence and Jira data. Companies like Clearwater Analytics have already seen productivity gains, demonstrating the effective collaboration between Codeium and Atlassian in optimizing the developer experience.

Latency is a significant constraint for AI assistant tools, particularly when using third-party large language models (LLMs) for autocomplete suggestions, as this can result in delays that affect usability, as evidenced by Sourcegraph Cody's experience. The latency issues arise from the complex process of collecting context, making model inferences, and merging results, which is exacerbated when relying on third-party APIs due to additional network delays and potential rate limiting. To address latency without altering model characteristics, developers have implemented solutions such as smart model compilation, optimized model architectures, model parallelism, and smart batching, which help reduce inference time by leveraging GPU resources more efficiently. These innovations allow for faster processing of larger models, enhancing the quality of AI-generated suggestions while accommodating a high number of concurrent users. The challenge is further compounded by the rise of open-source models, which tempt companies to build in-house solutions, often underestimating the necessary infrastructure and facing performance limitations.

Codeium has announced a partnership with MongoDB to enhance the development experience for MongoDB applications by integrating Codeium's AI assistant, which is specifically trained on MongoDB APIs and syntax. This collaboration aims to streamline the developer process by reducing the time spent searching through documentation and optimizing API usage through intelligent autocomplete suggestions. Developers can access these features by installing Codeium's extension, available for free on windsurf.com, which integrates seamlessly with IDEs like VSCode. The tutorial provided demonstrates setting up a movie review web app using NextJS and TailwindCSS with MongoDB as the data store, highlighting Codeium's capability to understand the codebase context and suggest efficient coding practices. The tool promises to improve code documentation and adherence to best practices while offering free individual access and additional tiers for team or enterprise use.

In its first year, Codeium focused on enhancing the IDE experience with advanced autocomplete and chat systems to facilitate code writing, but recognizing that software engineers engage in various tasks beyond coding, they expanded their AI capabilities to the terminal, resulting in Termium. Termium leverages the history of past commands and their outputs to offer autocomplete suggestions in the terminal, using the Tab key for acceptance or allowing users to continue typing or press Esc to ignore suggestions. The tool quickly garnered positive feedback from the Codeium team by simplifying tasks like source code management and analyzing Kubernetes logs, proving its utility in recalling commands and flags that typically require searching. Installation of Termium is straightforward, requiring users to run a command in their terminal and start a new session, with support currently available for MacOS and Linux.

As 2023 came to a close, Codeium introduced significant updates to enhance user experience, particularly in its chat functionalities within IDEs. Users can now utilize the @terminal feature to automatically integrate terminal history and outputs into chat messages, eliminating the need for manual copying. Enhancements include sticky model selection for GPT-4, allowing consistent model use across sessions, and customizable chat panel history options, enabling users to default to their most recent or a new conversation. Codeium's user base expanded over 300 times throughout 2023, and the company emphasizes the importance of user feedback in driving ongoing development as it looks forward to 2024.

Codeium enhances the development experience by integrating directly with your codebase parsing software, which is open source, and by utilizing @ mentions to automatically retrieve relevant context from a selection of popular open-source libraries. This feature aids in navigating large codebases with complex documentation and offers more accurate, up-to-date information. Additionally, Codeium integrates with the in-IDE terminal in VS Code to help parse large logs, understand complex outputs, and debug failed tests, ultimately boosting productivity. The developers have plans for further integration of Codeium into the development cycle and encourage feedback as they continue to release new features to enhance the Chat experience.

Characters per Opportunity (CPO) and Percentage Code Written (PCW) are proposed as definitive metrics to evaluate AI code assistants, with CPO focusing on benchmarking autocomplete capabilities and PCW assessing the actual value derived by developers using these tools. CPO measures the tool's effectiveness by evaluating suggestions without being subject to gameable factors, while PCW reflects the amount of code contributed by the assistant. Despite being conceptually related, CPO and PCW are not mathematically linked. Codeium uses these metrics to guide product development, boasting a CPO of 1.27 and a PCW of 44.6%, indicating that nearly half of the new code by its users is generated by the assistant. Autocomplete is emphasized as a critical feature due to its passive nature and the sheer volume of suggestions it can provide, significantly outweighing other features like chat in terms of value. The critique of competitors like SourceGraph Cody and TabNine highlights the importance of balancing acceptance rates with real value delivery, as optimizing solely for acceptance can lead to superficial improvements without enhancing the actual utility of the product. Codeium's focus remains on genuine product improvement and value delivery rather than chasing marketable metrics.