December 2025 Summaries

In 2025, the introduction of agentic systems significantly transformed data work by making complex workflows faster and more accessible to a diverse range of users beyond traditional data roles. This was not a year of minor improvements but a pivotal shift in how teams approached data work, driven by agents that could plan, reason, and execute tasks reliably. Zerve capitalized on this shift by emphasizing agentic systems and adopting a bottom-up, product-led growth strategy, allowing users to engage in real data work immediately without extensive sales processes. This approach attracted a wide variety of users, from experienced data scientists to RevOps and finance teams, who applied agentic technology in innovative ways. The company’s focus on user experience, such as integrating Canvas and Notebook views, providing block descriptions, and offering various agent modes, fostered a seamless and collaborative environment between humans and agents. These changes not only increased accessibility but also maintained the rigor of data work, expanding participation in meaningful data tasks more broadly than anticipated.

The text presents six essential data science techniques, each demonstrated through code examples available in public Zerve canvases for hands-on exploration. Techniques covered include one-hot encoding for transforming categorical data into numerical formats, groupby aggregations for summarizing data efficiently, and creating custom transformers to extend scikit-learn's functionality. It also discusses SQL-like window functions using pandas, the strategic use of custom loss functions to align machine learning models with business objectives, and the importance of vectorization in optimizing code performance for large datasets. Each section emphasizes practical coding patterns that data scientists routinely use, and the Zerve platform allows users to interact with, adapt, and integrate these examples into their workflows without requiring a login.

Zerve introduces reusable environments to streamline dependency management in data science by allowing users to build a set of dependencies once and apply it across multiple projects or workflows, eliminating the need for repeated setups and Docker image rebuilds. These reusable environments function as a ready-to-use toolbox for Python or R dependencies, which can be attached to entire projects, individual workflows, or specific blocks, ensuring consistent results and efficient collaboration. By facilitating cross-project reuse, specialized analyses, team collaboration, and versioning for experiments, Zerve enhances workflow efficiency and reliability, enabling teams to share consistent environments and save time, while also simplifying the onboarding process for new team members. This approach not only accelerates work by avoiding rebuild delays but also supports reproducibility and experimentation without impacting stable production workflows.

Dennis Oleksyuk, CTO and co-founder of AirCon, has developed an AI agent that autonomously manages 80% of air freight quotes while freight forwarders are asleep, highlighting the inefficacy of many agentic frameworks in production settings. During a livestream discussion, Oleksyuk emphasized that popular frameworks are often unsuitable for production due to their focus on rapid prototyping rather than reliability. He advocates for using raw LLM APIs for greater control and adaptability, especially when dealing with the complexities and corner cases inherent in real-world business processes. Oleksyuk's AI solution not only automates quoting but also manages bookings across multiple providers, enabling freight forwarders to generate revenue effortlessly. The key to success, he argues, lies in deeply understanding the domain to effectively manage exceptions rather than relying on trendy frameworks. This approach has allowed AirCon to maintain a competitive edge by cataloging and solving numerous corner cases, ultimately transforming a traditionally manual process into an efficient, automated system.

Zerve has introduced an innovative notebook interface that reimagines how teams can collaboratively engage with data workflows by integrating AI capabilities specific to data tasks. Unlike traditional tools like Jupyter or Google Colab, Zerve's platform operates entirely in the cloud, allowing seamless collaboration akin to Google Docs, and incorporates an AI agent that understands data processes to automate and expedite exploratory analysis and other tasks. During a demonstration, the system showcased its capacity to transform code into apps, deploy APIs, and execute parallel processing, significantly reducing the time required for experimentation and increasing the efficiency of data science workflows. The creators highlighted the platform's security features, such as easy self-hosting, which allows organizations to control their data in their own infrastructure, making it suitable for robust enterprise use. Looking forward, Zerve aims to evolve its AI from a reactive tool to a proactive partner in data science, with potential future capabilities including autonomous problem-solving and predictive suggestions, signaling a transformative shift in how data professionals might engage with their analytical tools.

Zerve Notebook is an advanced data workspace that integrates a familiar notebook interface with the Zerve agent, enabling users of all skill levels to conduct sophisticated data analysis, model building, and deployment within a unified platform. This system enhances traditional notebook workflows by providing real-time collaboration, stable execution models, and an intuitive, agent-assisted environment that helps with coding and decision-making. The Zerve Notebook allows for seamless transition from exploration to deployment, supporting parallel execution and visual workflows, while maintaining clarity and reproducibility through a DAG-based architecture. Users can interact with data using natural language and code, with the Zerve agent acting as a collaborative partner to streamline processes, ensure reliability, and facilitate understanding of complex data projects. The platform's design eliminates common issues associated with traditional notebooks, such as execution unpredictability and collaboration friction, allowing for quicker transitions from ideas to production-ready systems.