The Death of Data Gatekeeping: AI Makes Everyone An Analyst | Hex Cofounder

Key Takeaways

AI is transforming data analysis by providing direct answers and contextual insights beyond traditional dashboards.
Hex leverages AI agents and conversational interfaces like Threads to democratize data access and accelerate analysis for all users.
Ensuring trustworthiness in AI-generated data insights requires strong context engineering and semantic modeling.
Data teams are evolving, with new roles focused on AI governance, context management, and deep-dive analytical support.
Hex integrated AI as a core product component, dissolving its dedicated AI team to embed AI capabilities across all features.
The trend in SaaS is to make AI fundamental, influencing pricing models and emphasizing value capture over separate AI charges.
The "post-modern data stack" emphasizes open formats and modular query engines for greater data sovereignty.
Palantir's original "forward-deployed engineering" model involved product engineers building custom solutions on customer sites.
"Commitment engineering" at Hex secures incremental customer buy-in, de-risking sales and product development.

Traditional dashboards often fail to provide direct answers, leading to low data-informed decision rates.
Historically, data analysis workflow was fragmented with separate SQL, CSV, and notebooks, hindering collaboration.
Hex initially aimed to unify this, but AI now acts as an analytical partner, significantly accelerating question-answering.
AI broadens access to data analysis within organizations by simplifying query processes and reducing friction.

Hex's agentic capabilities, like the notebook agent and Threads, advanced due to improved AI model reasoning and iteration since 2021.
Recent language models, including Gemini 3.0, can reason, iterate, and engage in multi-turn conversations, enabling complex AI-driven analysis.
Data analytics involves hundreds of thousands of tokens for query results or chart images, posing context window challenges for AI agents.
Hex collaborates with Anthropic and OpenAI to address large context volumes and ensure both speed and expanded context capabilities.

Context engineering is crucial for great AI data products, particularly semantic models that instruct AI on complex database structures.
Semantic models help AI reason about data, especially when direct database interaction is insufficient for accurate answers.
Hex integrates with tools like DBT and offers an in-product agent to create semantic models as part of a larger context feedback loop.
Trustworthiness, not just conversational interfaces, is paramount for data professionals, as incorrect answers quickly erode user trust.
Hex develops toolkits to monitor user questions and evaluate answer quality, using agents to review responses for satisfaction and model confusion.

The role of data teams is evolving to include significant AI governance responsibilities.
Hex's data team, including former customers, is overprovisioned but plays a crucial role in dog-fooding new features and defining the future of data analysis.
Their evolving responsibilities focus on deep-dive analyses and context engineering, acting as a "spider on a web" to detect emerging questions.
The concept of accuracy for AI systems is complex; "trustworthiness" is preferred for stochastic, multi-turn models, especially in mission-critical applications.

Hex began experimenting with AI with GPT-2 and launched initial AI features shortly after ChatGPT's release, branding them as "magic."
In early 2023, Hex dissolved its dedicated AI team, recognizing that agentic models were maturing and AI was becoming fundamental to the product.
This organizational shift empowers all product teams to integrate AI, aligning with the broader trend of AI becoming core to all software.
Building with large language models (LLMs) is akin to research, requiring deep understanding of evolving model capabilities and relationships with AI labs.

Hex does not charge separately for AI features, viewing AI as core to the product, analogous to charging for cloud services in 2014.
While SaaS companies initially offered AI as an add-on, the sentiment is shifting towards integrating AI as a core product value, a move Hex embraced early.
Increased token usage in agent features may lead to consumption-based pricing or overages for heavy users, though AI will remain a core offering.
Hex mandates AI features for new customers to ensure alignment with the product's direction and a better user experience, even if it impacts short-term revenue.

The modern data stack, once a novel concept popularized by companies like DBT and Fivetran, has become an industry standard.
The "post-modern data stack" emphasizes storing data in commodity storage like S3 using open formats such as Iceberg.
This new architecture allows for modular query engines and greater data sovereignty, potentially reducing costs associated with cloud data warehouse monopolies.
Hex acquired Hashboard, a business intelligence tool, as a strategic bet on owning the entire data insight layer, encompassing exploration, analysis, and sharing.

Barry McCardel clarified his tweet about "forward-deployed engineers" (FDEs) was a playful critique emphasizing Palantir's authentic FDE model.
Palantir's original FDE model involved sending product engineers directly to customer sites, including challenging environments, to deeply understand and solve problems.
These engineers built custom solutions that informed the development of the core platform, with successful solutions integrated back into the main product.
True FDE involves engineers writing production code in the field, not just configuration or short-term customization.

Hex adopted "commitment engineering," inspired by Palantir, to secure incremental commitments from customers throughout the product development and sales process.
This strategy involves progressing from simple conversations to pilots and trials to gauge genuine interest before final payment.
It helps de-risk the sales cycle and product development by seeking truth through customer commitments rather than just positive feedback.
This approach aims to prevent startups from being led astray by superficial endorsements.