Barr Yaron: AI Engineers Value Quality Over Open Models

Barr Yaron, Partner at Amplify, presented the findings of the "2026 State of AI Engineering Survey," highlighting key trends and predictions for the AI development landscape. Speaking at the AI Engineer World's Fair, Yaron emphasized that despite the buzz around open-weight models, AI engineers are primarily driven by model quality and agentic capabilities, with cost also emerging as a critical factor.

Key Survey Findings on Modalities and Models

The survey revealed that text continues to dominate the modalities AI engineers are building with. However, image generation saw a significant leap, with the percentage of engineers using it and feeling positive about its performance doubling from 18% last year to 36% this year. Audio, while not showing the same breakout growth, maintains the strongest intent-to-adopt rate, with 56% of those not currently using it planning to do so.

Regarding models, Yaron noted that while 94% of respondents use closed models, open-weight models are increasingly being adopted, with 45% of users also leveraging closed models. This suggests a trend of augmentation rather than replacement, with teams mixing and matching to suit their needs. Crucially, the debate between open and closed models is secondary to core considerations like quality, agentic capabilities, and cost, which were cited as top priorities by the vast majority of respondents.

The Rise of Agents and the Impact of Cost

The survey also highlighted the significant growth in agent adoption, with 95% of teams now using them, up from roughly half last year. A key shift observed is the increase in agents having write access, with 89% now capable of writing data, up from 52% last year. This move from read-only functions to action-taking agents raises questions about control mechanisms, with human-in-the-loop approvals and gating permissions being the most common, albeit described as "blunt instruments." Hallucinations and context loss remain the primary frustrations with agent performance.

The financial aspect of AI is also becoming a major consideration. A staggering 76% of respondents stated that cost regularly or sometimes shapes their AI usage ambition. This has led to cost becoming a "first-class engineering constraint," influencing product decisions and being monitored as closely as quality itself.

Organizational Shifts and Future Outlook

AI's impact on teams is largely positive, with 97% of respondents reporting a net positive effect, primarily driven by increased experimentation and faster shipping cycles. However, concerns about the erosion of deep technical skills due to AI-generated code are present. The lines between roles are blurring, with 81% of engineers feeling their roles are expanding into product design and marketing. Notably, over a third of teams now have non-developers shipping features, indicating a democratization of software development.

Looking ahead, 67% of respondents expect a declared AGI within five years, and most anticipate AI-generated research. The survey also revealed a split in opinions regarding AI compute in space versus on land, and a significant portion of engineers feel current AI code reviews don't scale, highlighting the need for evolving evaluation methods.