April 2026 Summaries

Hex has developed a visualization subagent to enhance its main AI agent's ability to handle visualization tasks independently, improving the quality of visual outputs and allowing the main agent to focus on broader analytical tasks. By offloading visualization-specific tools to this specialized subagent, the main agent is relieved from the complexity and token usage associated with visualization intricacies. The subagent is designed to iteratively refine visualizations by focusing first on foundational aspects such as data structure and axes before moving on to styling. This approach is informed by visualization best practices and the "agent accessibility principle," ensuring the agent perceives and refines its outputs effectively. Evaluation of the subagent's performance is done using a modified version of Microsoft's VisEval benchmark to ensure that it not only executes visualization tasks but also makes qualitative judgments, while bug-specific tests help refine and expand its capabilities. This iterative cycle of development and evaluation aims to continually improve the agent's visualization proficiency, with Hex inviting users and potential employees to engage with and contribute to this evolving technology.

In the April 2026 edition of "Hex in the Wild," several professionals share how they utilize Hex to enhance operations across different sectors. Léo Tavernier, a Senior Data Scientist at Gojob, developed a Hex application to validate social security ID cards, reducing processing costs by 75% by leveraging Hex's Python integration and dynamic parameters. Jiaying Du, an Analytics Engineer Lead at Profound, created a customer health dashboard in Hex to predict churn and expansion, utilizing SQL queries and dynamic filters without needing engineering support. Kevin Sloan, a Growth Marketing Manager at Descript, used Hex to build a performance dashboard that analyzes affiliate marketing data, identifying key trends and opportunities among partners. These case studies illustrate Hex's capacity to empower non-engineering teams to build data-driven solutions efficiently.

Amanda Fioritto details the Hex data team's journey to automating data validation processes, culminating in the development of an AI-powered audit-helper skill that significantly enhances the efficiency and accuracy of their validation workflows. The text describes how the team initially used dbt's audit_helper package for standardized testing but later sought more flexibility and visibility, leading them to build the Hex Audit Helper app. As AI tools advanced, they developed an agent skill to conduct end-to-end audits with principles of standardization, flexibility, and traceability, ensuring that validation remains a human-guided process despite automation. The audit-helper skill incorporates explicit guidelines to handle large models, reduce noise, and avoid assumptions by requiring evidence for changes, thereby making validation quicker and more comprehensive. The text emphasizes the role of AI as a supportive partner rather than a replacement for human judgment, aiming to make data validation more seamless and impactful, ultimately enabling the data team to focus on higher-value tasks and improve decision-making across the organization.

Barry McCardel's piece discusses the concept of "Context Exhaust," a term used to describe the automatic generation of valuable context signals during everyday software usage, particularly with AI systems. He argues that while the buzz around context in AI is widespread, the ways it is built and leveraged remain unclear. Hex, the company he mentions, focuses on capturing and refining these signals into semantic models and guides, thereby creating a self-improving feedback loop that enhances data accuracy over time. This process allows for the integration of external data sources and the refinement of context through user interactions, ultimately fostering a collective learning environment within organizations. The article emphasizes that becoming "data driven" is not about isolated chart analysis but involves a collaborative, iterative, and evolutionary approach where organization-wide knowledge is continuously fed and improved by context exhaust.

Hex is introducing Context Suggestions, an innovative system designed to improve data analysis by creating a self-learning environment that becomes more intelligent with each use. This system leverages automated context improvements generated from ongoing user interactions and analyses, allowing for a more scalable and automated approach to data inquiries. By utilizing Data Agents and Review Agents, Hex automatically learns from user conversations, identifying context gaps and suggesting specific changes to enhance understanding and responses. This process not only surfaces emergent data topics but also flags areas for improvement, enabling data teams to make informed updates to guides, semantic models, and metadata. The platform is built to foster a connected ecosystem where all data analyses and insights are accessible across various interfaces, ultimately enhancing the organization's ability to conduct impactful and integrated data analysis.

Model drift refers to the decline in predictive accuracy of a model post-deployment due to changing conditions, often manifesting as data drift, concept drift, prediction drift, training-serving skew, or upstream drift. Most drift issues are traced back to data quality or upstream pipeline problems rather than the model itself, necessitating distinct remedies depending on the type of drift. For instance, data drift involves changes in input distribution, while concept drift occurs when the relationship between inputs and outputs shifts. Upstream drift, often the most common, results from changes in the data pipeline rather than real-world changes. Effective drift detection involves maintaining a baseline of training data statistics and employing statistical checks like the Population Stability Index (PSI) to monitor shifts, while governance and observability practices, such as centralized metric ownership and version control, enhance model reliability. Addressing drift requires diagnosing the root cause before deciding on actions like retraining, which should be considered after verifying performance issues and data quality. To mitigate drift, it's essential to focus on pipeline integrity, metric definitions, feature engineering, and the use of shared, endorsed data definitions to ensure consistency and transparency in analytics workflows.

Olivia Koshy addresses a future Product Manager at Hex, highlighting the evolving landscape of data and the company’s pivotal role in it. Hex, initially built to streamline fragmented data workflows, is now well-suited for the current era where AI impacts every organizational level. The company offers a robust platform featuring a graph-based execution system, sandboxed cloud environments, and a warehouse-agnostic querying system, which are essential for AI to effectively analyze organizational data. Hex is trusted by leading AI-driven companies and is experiencing significant adoption of agent technologies. Upcoming projects include improving code generation, enhancing dashboards, and developing autonomous agents for proactive organizational insights. Olivia emphasizes the need for innovative thinkers who challenge existing roadmaps and are passionate about quality and detail. She encourages those who are excited by these opportunities and the transformative power of data to join Hex, which is actively expanding its team.

AI sprawl, characterized by inconsistent outputs from different AI tools within an organization, arises from structural issues where each tool operates in isolation without a centralized, trusted data context. This leads to conflicting answers, particularly apparent in business meetings where teams spend time verifying data instead of making decisions. The Hex Team underscores the importance of establishing a single, governed layer of business definitions that all AI systems can access, ensuring consistency and accuracy in outputs. They suggest a gradual, seven-step approach to manage AI sprawl, emphasizing the need for shared context over merely adding new tools. By consolidating AI and data governance under a unified architecture, organizations can reduce the hidden costs associated with AI sprawl, such as wasted analyst hours and technical debt. The approach focuses on improving metric definitions, enhancing AI behavior observability, and ensuring every AI-generated answer derives from a single source of truth, ultimately allowing teams to focus more on meaningful work.

Hex has introduced a command-line interface (CLI) designed to streamline data team operations by integrating seamlessly into existing terminal workflows, allowing data teams and coding agents to programmatically interact with Hex without disrupting their current processes. This shift reflects the evolving nature of data work where teams increasingly rely on coding agents and sometimes voice-to-text technology to instruct these agents. The Hex CLI facilitates the creation and management of projects, cells, and SQL or Python updates directly from the terminal, promoting efficiency and reducing the need to switch between different tools. It also supports the integration of workflows and allows for large-scale changes across projects, such as updating SQL style guides or data connections. The CLI’s launch aims to make Hex a central hub for data work, with further developments anticipated as users explore and suggest new applications for its capabilities.

As organizations increasingly prioritize AI adoption, they face the challenge of ensuring AI outputs are trustworthy, a concern highlighted by Hex's State of Data Teams 2026 report, which notes a significant increase in data leaders citing AI as a top goal, with trust being a primary concern for 31% of them. The adoption of AI is often hampered by treating it as a mere tooling decision rather than addressing the context in which it operates, leading to AI systems providing confident but potentially incorrect answers due to a lack of understanding of data specifics. To overcome this, companies are advised to focus on establishing a robust context infrastructure by endorsing tables, adding descriptions, and implementing workspace rules, with semantic models as a long-term goal rather than a prerequisite. A phased adoption strategy is recommended, starting with controlled pilots to refine context and gradually expanding access, ensuring AI is deployed effectively across different analytics tasks by matching tools to the nature of work rather than specific roles. Trust in AI builds over time through visibility and the ability to verify AI outputs, rather than being a precondition, with governance infrastructure playing a crucial role in enabling continuous improvement and successful AI deployment.

In the realm of AI analytics, context engineering is emphasized as a crucial practice for ensuring that AI systems produce accurate and reproducible results, rather than relying solely on prompt engineering. While prompts determine how to ask questions, context engineering focuses on what the AI knows prior to answering, involving the endorsement of trustworthy tables, creation of workspace rules, and development of semantic models. This structured environment allows AI to accurately interpret data, reducing errors such as inconsistent metrics, incorrect joins, and confidently wrong answers. Hex's State of Data Teams 2026 report highlights trust as a primary concern in AI adoption, with context engineering identified as a key solution to enhance AI's reliability. The process of building context is iterative, involving layers from endorsed tables to observability mechanisms, allowing teams to progressively deepen their AI's understanding and improve data-driven decision-making. By prioritizing context over prompts, data teams can ensure more consistent and trustworthy outcomes, making AI analytics more reliable and actionable.

In an era where artificial intelligence increasingly automates predictable, routine tasks in line with its training data, the importance of human creativity and unconventional thinking is emphasized in a memo by Barry McCardel to the team at Hex, advocating for the recruitment of individuals who think and act "out of distribution." These "weirdos," who are characterized by their unusual, innovative, and sometimes implausible ideas, are seen as vital to advancing the company and pushing boundaries beyond AI's capabilities. McCardel argues that as organizations grow, they must resist becoming overly conservative and instead focus on hiring and valuing those who bring unique perspectives and break the mold. This approach, he suggests, is crucial for continuing Hex's success and fostering a culture of originality and innovation that leverages human potential to complement AI's limitations.

In the realm of AI analytics, the distinction between prompt engineering and context engineering is crucial, especially when ensuring the accuracy and trustworthiness of analytical outputs. Prompt engineering involves refining communication with AI models to improve outputs, suitable for tasks like summarizing datasets or reformatting queries. However, as AI use scales within organizations, context engineering becomes vital; it involves designing the information environment the AI operates in, including schema metadata, metric definitions, and governance rules. This approach addresses the limitations of prompt engineering by providing a structured framework that ensures consistent, defensible answers across complex queries, such as accurately calculating revenue data from vast enterprise databases. By implementing context engineering, organizations can prevent errors like incorrect joins, inconsistent metric definitions, and unauditable results, which are common when AI operates without governed context. The progression from endorsing tables and adding descriptions to building semantic models and observability loops exemplifies how context engineering can be integrated into existing semantic layer tools, enhancing AI's reliability and fostering trust in AI-driven analytics.

Hex has launched Pure Glass, a transformative design system aimed at redefining user experience by making interfaces more expressive, transparent, and engaging. This innovative system utilizes a translucent material that allows users to see through the interface, thereby reducing visual clutter and enhancing clarity by blending buttons and menus with underlying content. The redesign extends to all parts of the Hex product, including visualizations, ensuring that insights are as transparent as the interface itself. Hex's leadership emphasizes their commitment to pushing design boundaries and believes this bold new direction will significantly enhance the platform, which is designed to facilitate the creation and sharing of interactive data products. The company invites users to experience Pure Glass and explore opportunities to join their team.