{"id":13673,"date":"2025-05-30T14:43:10","date_gmt":"2025-05-30T22:43:10","guid":{"rendered":"https:\/\/live-cometml.pantheonsite.io\/?page_id=13673"},"modified":"2026-01-12T20:11:15","modified_gmt":"2026-01-12T20:11:15","slug":"automatic-prompt-optimization","status":"publish","type":"page","link":"https:\/\/www.comet.com\/site\/products\/opik\/features\/automatic-prompt-optimization\/","title":{"rendered":"Agent Optimizer"},"content":{"rendered":"\n
\n

Automate Prompt Engineering
& Agent Optimization<\/h1>\n\n\n\n

Step up your agent development workflows using Opik to automatically iterate and tune system prompts you can freeze and deploy in production.

Unlock cost, performance, and reliability improvements in a fraction of the time you\u2019d spend on manual prompt engineering.<\/p>\n\n\n\n

\n
GitHub<\/a><\/div>\n\n\n\n
Free Cloud Version<\/a><\/div>\n<\/div>\n\n\n\n
\n

Trusted by the most innovative AI teams<\/p>\n\n\n\n

\n\t
\n\n
\n
\"AssemblyAI<\/figure>\n\n\n\n
\"Natwest<\/figure>\n\n\n\n
\"Stellantis<\/figure>\n\n\n\n
\"Uber<\/figure>\n\n\n\n
\"zencoder<\/figure>\n\n\n\n
\"Netflix<\/figure>\n\n\n\n
\"Autodesk<\/figure>\n\n\n\n
\"Etsy<\/figure>\n\n\n\n
\"Stability<\/figure>\n\n\n\n
\"Mobileye<\/figure>\n<\/div>\n\n<\/div>\n\t\n\t
\n\n
\n
\"AssemblyAI<\/figure>\n\n\n\n
\"Natwest<\/figure>\n\n\n\n
\"Stellantis<\/figure>\n\n\n\n
\"Uber<\/figure>\n\n\n\n
\"zencoder<\/figure>\n\n\n\n
\"Netflix<\/figure>\n\n\n\n
\"Autodesk<\/figure>\n\n\n\n
\"Etsy<\/figure>\n\n\n\n
\"Stability<\/figure>\n\n\n\n
\"Mobileye<\/figure>\n<\/div>\n\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n\n\n\n
\n
\n
\n

Automate Multi-Trial Optimization at Scale<\/h2>\n\n\n\n

Automated prompt engineering<\/a> gives your complex agentic systems the most predictable LLM performance across multiple models, with efficient iteration built into your LLM eval lifecycle.<\/p>\n<\/div>\n\n\n\n

\n
\"flow<\/figure>\n<\/div>\n<\/div>\n\n\n\n
\n
\n

Generate, Score, & Implement Elite Prompts<\/h2>\n\n\n\n

The Opik Agent Optimizer SDK comes with everything you need to generate new prompts automatically, score them based on your eval criteria, and trust that the best version makes it to production.<\/p>\n<\/div>\n\n\n\n

\n
\"code<\/figure>\n<\/div>\n<\/div>\n<\/div>\n\n\n\n
\n

Run Seven<\/mark> Prompt Optimization Algorithms in One Package<\/h2>\n\n\n\n
\n
\n

Few-Shot Bayesian
Optimizer<\/h3>\n\n\n\n