Beyond the Benchmark: Why the Harness Matters

Arito’s verified Spreadsheet Bench v2 result shows why frontier models alone aren’t enough. A purpose-built AI agent harness helps turn spreadsheets into structured business knowledge for Arito’s semantic layer and knowledge graph.

Why the Agent Harness Matters More Than the Model?

It’s hard to believe that our name is above OpenAI, Anthropic and Google.

Did we train a completely new model? No. Well…we did initially but then we changed our approach once we learned what actually matters.

Our verified SpreadsheetBench v2 result suggests otherwise.

Thing is… we use frontier models too, but what makes the difference is the harness around them: spreadsheet-specific tools, visual feedback, Excel’s real calculation engine, context management and verification.


𝐓𝐡𝐞 𝐦𝐨𝐝𝐞𝐥 𝐩𝐫𝐨𝐯𝐢𝐝𝐞𝐬 𝐜𝐚𝐩𝐚𝐛𝐢𝐥𝐢𝐭𝐲. 𝐓𝐡𝐞 𝐡𝐚𝐫𝐧𝐞𝐬𝐬 𝐦𝐚𝐤𝐞𝐬 𝐢𝐭 𝐮𝐬𝐚𝐛𝐥𝐞.

And there’s a bigger reason this matters to Arito: Spreadsheets are basically executable business knowledge: forecasts, pricing logic, unit economics, assumptions and operating models. 𝐈𝐟 𝐰𝐞 𝐜𝐚𝐧 𝐝𝐞𝐞𝐩𝐥𝐲 𝐮𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝 𝐭𝐡𝐞𝐦, 𝐀𝐫𝐢𝐭𝐨 𝐜𝐚𝐧 𝐥𝐞𝐚𝐫𝐧 𝐡𝐨𝐰 𝐚 𝐜𝐮𝐬𝐭𝐨𝐦𝐞𝐫’𝐬 𝐛𝐮𝐬𝐢𝐧𝐞𝐬𝐬 𝐰𝐨𝐫𝐤𝐬 𝐦𝐮𝐜𝐡 𝐟𝐚𝐬𝐭𝐞𝐫.

Spreadsheets then become more than files to edit. They become a 𝐟𝐢𝐫𝐬𝐭-𝐜𝐥𝐚𝐬𝐬 𝐬𝐨𝐮𝐫𝐜𝐞 𝐟𝐨𝐫 𝐨𝐮𝐫 𝐬𝐞𝐦𝐚𝐧𝐭𝐢𝐜 𝐥𝐚𝐲𝐞𝐫 𝐚𝐧𝐝 𝐤𝐧𝐨𝐰𝐥𝐞𝐝𝐠𝐞 𝐠𝐫𝐚𝐩𝐡 – alongside CRM, ERP and warehouse data.

The ranking is great validation. But the capability underneath it is what really matters.

Now….back to building 🚀

TRY ARITO NOW

Ready for continuous identity protection without gaps or guesswork?

About The Author

Daniel Zahavi

Founder | CEO

Daniel Zahavi is CEO of arito. He is a second-time founder and product builder with a PhD in information theory and deep expertise in data modeling and solving complex data problems. He previously founded a company focused on data infrastructure and analytics, which was successfully exited; its technology is now deployed widely across the United States. Today, he focuses on building AI-powered products and improving AI efficiency, translating cutting-edge models into reliable, scalable systems that automate workflows and deliver real business outcomes, drawing on years of hands-on experience defining, shipping and iterating products end-to-end.