← back
The 2025 AI Engineering Report — Barr Yaron, Amplify
Takeaway
AI engineering in 2025 is heterogeneous, evaluation-bottlenecked, and dominated by RAG plus a surprising amount of fine-tuning.
Summary
- Amplify's 2025 survey of 500 AI engineers: half of seasoned 10+yr software engineers have ≤3 years AI experience; the field is newly multidisciplinary.
- 94% of LLM users use them for 2+ use cases, 82% for 3+; OpenAI provides 3 of the top 5 customer-facing production models.
- 70% use RAG; fine-tuning adoption higher than expected with 40% using LoRA/QLoRA, plus DPO, RFT, and supervised SFT — many hybrid approaches.
- 50%+ update models monthly, 17% weekly; 70% update prompts monthly, 10% daily; 31% have no prompt management at all.
- Only <20% say agents work well at work (vs 80% for LLMs); 65% use dedicated vector DBs; biggest pain point cited: evaluation.
ai-engineeringsurveyevals
Original description
Come hear the results of the 2025 State of AI Engineering: https://www.amplifypartners.com/blog-posts/the-2025-ai-engineering-report About Barr Yaon Barr is a data scientist turned investment partner at Amplify Partners where she invests in AI infrastructure and apps Recorded at the AI Engineer World's Fair in San Francisco. Stay up to date on our upcoming events and content by joining our newsletter here: https://www.ai.engineer/newsletter