/Tech3h ago

Baturay Saglam and Cisco's Amin Karbasi release FAPO to automate agent prompt tuning, beating GEPA on 15 of 18 benchmarks

It achieved a 14.1 percentage point average performance increase.

4222153.2K

#873

Original post

Foundation AI@fdtn_ai

Introducing FAPO: Fully Automated Prompt Optimization.

As LLM systems become increasingly agentic, manually tuning prompts across complex workflows quickly becomes a bottleneck. FAPO autonomously optimizes multi-step LLM pipelines—including ReAct agents operating over MCP—with optimization runtime powered by Claude Code and Codex.

In work led by AI Research Intern, Baturay Saglam, FAPO outperforms GEPA, the current SOTA prompt optimizer, on 15 of 18 benchmarks, delivering an average improvement of +14.1 percentage points.

Check out the blog from our AI researcher Huaibo Zhao in the comment.

8:53 AM · Jun 19, 2026 · 2.1K Views

Sentiment

Sentiment building, check back later.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS1.3KBOOKMARKS3LIKES7RETWEETS1REPLIES2

Amin Karbasi@aminkarbasi

Friends don’t let friends write prompts by hand, they give them FAPO.

Foundation AI@fdtn_ai

Introducing FAPO: Fully Automated Prompt Optimization.

In work led by AI Research Intern, Baturay Saglam, FAPO outperforms GEPA, the current SOTA prompt optimizer, on 15 of 18 benchmarks, delivering an average improvement of +14.1 percentage points.

Check out the blog from our AI researcher Huaibo Zhao in the comment.

2h1.3K73

Soheil Feizi@FeiziSoheil

@aminkarbasi Great work @aminkarbasi

Amin Karbasi@aminkarbasi

Friends don’t let friends write prompts by hand, they give them FAPO.

37m4910

Foundation AI@fdtn_ai

Blog Post Link: https://cisco-foundation-ai.github.io/blogs/fully-automated-prompt-optimization/ Technical Report Link: https://arxiv.org/pdf/2606.19605v1

3h10