Before you build that data agent, I have notes.

This post provides a critical perspective on building and owning data agents, emphasizing the significant ongoing costs and complexities beyond initial prototyping. It highlights three key areas of ownership: auditability and governance for "is this right?" questions, continuous upkeep of context and business logic as companies evolve, and the expansion of agent capabilities beyond chat to full analytics platforms. The author, drawing from experience at Vercel, Miro, and Google, argues that while prototyping is easy, the long-term maintenance, debugging, and feature expansion of a production data agent represent substantial engineering effort. The post advocates for a careful build vs. buy analysis, suggesting that for most companies, the unique business logic and trusted models are the core assets worth owning, rather than the entire agent infrastructure. It also introduces a calculator to help teams estimate the true cost of building their own agents.