Barr Yaron presents the 2026 State of AI Engineering survey, conducted with Notion and Vercel and answered by 1,048 practitioners. The sample spans engineers, founders and product roles and skews toward builders; its percentages describe those respondents, not the entire workforce. Text remains dominant, while image generation gained satisfied users and audio showed strong intended adoption.
Respondents commonly combine closed and open-weight models rather than choosing one camp. Yaron distinguishes standardization of platforms and tools from standardization on a single model. Cost constrains product ambitions and is monitored alongside task quality, while agents increasingly have permission to write data without a settled approach to control.
Evaluation remains a leading stack challenge. Teams often buy inference infrastructure but keep prompts and product logic in-house. Cheaper experimentation and broader participation in shipping coexist with heavier review burdens, weaker codebase understanding and concern about long-term maintenance. Predictions about AGI declarations and future architectures are survey opinions, not demonstrated outcomes.
Watch on YouTube
