Dataiku Executive Outlines Enterprise Agent Strategy

At the Ai4 2026 conference in Las Vegas, Jed Dougherty, senior vice president of AI and platform at Dataiku, outlined how enterprises can navigate the rapid evolution of generative and agentic AI. Although adoption has lagged behind technology development since OpenAI released Chat-GPT in 2022, recent breakthroughs have spurred interest. These include the OpenClaw open-source personal agent, Anthropic's Claude Cowork and Mythos models, and Nvidia CEO Jensen Huang's call for an "OpenClaw strategy." Dougherty emphasized that despite these autonomous advances, the most successful enterprises still keep humans in the loop.

To bridge the technical gap, Dougherty highlighted the role of forward-deployed engineers in helping clients overcome infrastructure hurdles. He explained that while users were comfortable with self-service machine learning four years ago after practicing it for 10 years, the complexity of modern agentic pipelines requires expert guidance. For enterprises looking to deploy these technologies, Dougherty advised focusing on core business processes. For example, an insurance company should map its claim processing, investigation, and auditing workflows to identify which specific subsets can be augmented by agentic AI.

Managing the rising costs of AI is another critical challenge, with Dougherty predicting a wave of "sticker shock" by 2027 similar to the early days of cloud computing. He admitted to personally burning through $1,000 in tokens a day during intensive work. To mitigate these expenses, he advised enterprises to break agent tasks into small steps and remain open to switching between foundation models based on price and capability. This includes utilizing competitive open-source options from Chinese providers like Alibaba and DeepSeek.

Ultimately, Dougherty recommended that enterprises establish a centralized, agnostic orchestration platform. This layer allows organizations to safely mix, match, and swap underlying tools, whether utilizing Snowflake, OpenAI, or on-premises models for sensitive data. By maintaining this flexibility, practitioners can easily choose the right sub-tool for the right task while keeping operational costs under control.