The fight to tame spreadsheets with AI isn’t over yet. A new company called Meridian has emerged from stealth with a more comprehensive IDE-based approach to agentic financial modeling — and plenty of funding to build it. On Wednesday, the company announced $17 million in seed funding at a $100 million post-money valuation.
“Our goal is to make financial modeling and spreadsheets way more predictable and auditable,” CEO and co-founder John Ling told TechCrunch. “How can you take a process that traditionally might have taken several hours and condense it down into like 10 minutes?”
The round was led by Andreessen Horowitz and The General Partnership, with participation from QED Investors, FPV Ventures, and Litquidity Ventures. The company says it is currently working with teams at Decagon and OffDeal, and signed $5 million of contracts in December alone.
Excel agents have been a popular target for AI startups, due in part to the high cost of human-led financial analysis. But where previous Excel agents like Shortcut AI built agents into Excel, Meridian operates as a stand-alone workspace, more akin to Cursor. This allows the app to operate like an IDE, integrating data sources and other outside references that might otherwise create friction.
Based in New York, the Meridian team includes both alumni of AI firms like Scale AI and Anthropic as well as financial veterans from firms like Goldman Sachs.
As Ling describes it, Meridian’s biggest challenge is the strict requirements of financial clients, which often clash with the non-deterministic nature of AI models.
“If you go to 10 different software engineers at Google, and you want to add some new feature into an app, you’ll probably get like 10 completely different implementations. And that’s totally fine,” Ling says. “But if you go to 10 banking analysts at Goldman Sachs and you ask for 10 valuation models for a company, you would probably get 10 almost identical workbooks.”
Techcrunch event
Boston, MA
|
June 23, 2026
As a result, the Meridian team has done significant work to make their outputs more auditable and deterministic, while maintaining the flexibility of LLM-based tools. The result is a mixture of agentic AI and more conventional tooling, minimizing the hallucinations that slow down many enterprise deployments.
“Our goal is to really remove the doubt layer right from the LLM process,” Ling says. “You know exactly how the logic flows, and all of these assumptions or whatever that go into the model, you can see exactly where they’re coming from.”
