Three releases on November 5, from insurance, banking and manufacturing, all landed on the same design for AI agents: let them do the work, keep a person on the decision.
Kalepa, a New York underwriting software company, raised a $75 million Series B at a $470 million valuation, led by Updata Partners with IA Ventures and Inspired Capital joining. Inspired led its $14 million Series A in 2021, so the round lands five years after the last one. Kalepa says it’s been cash flow positive for more than two years and has raised $91 million in total.
The product coordinates AI agents through multi-step underwriting, pricing and portfolio work, and brings in an underwriter for judgment or approval. Every output is traceable. Kalepa says insurers using it see more than 58% higher premium per underwriter, quotes 30% to 60% faster and a combined ratio 960 basis points better. Co-founder Paul Monasterio described the split plainly: sometimes the right answer comes from AI, sometimes from a person. The money goes to new markets and doubling the team by 2027. Competitors raised too this year: Federato $100 million, Sixfold $30 million, Artificial Labs $45 million.
Banks are piloting, few are orchestrating
Capgemini’s World Cloud Report for Financial Services surveyed 1,100 executives across 14 markets in June and July. 59% are piloting or running multi-agent AI. Only one in ten has the orchestration to run those systems on their own across the enterprise, and just 5% have made agentic AI an enterprise-scale capability. About half can’t consistently show a return on their agent projects.
The plumbing is where the spending goes next. 90% use or plan to use model routing to control token costs, and 52% want hyperscalers to govern their agent deployments. Three quarters expect agents to be central to customer engagement within three years, mainly in account servicing, loan origination, underwriting and product recommendations.
Factories trust the output less than the tool
Rootstock’s survey of 588 manufacturers in the US, UK and Canada found 77% plan to raise AI budgets over the next 12 to 18 months, down from 82% a year earlier. 31% of those deploying AI already run agentic AI. The barrier has changed. Skills, integration and cost were the 2024 complaints. Now, Rootstock’s product chief Ohad Idan said, the top concern is whether the output can be trusted, and an AI that’s wrong one time in a thousand can be more dangerous than one that’s wrong often, because people stop checking it.
So 73% of manufacturers want AI that’s directed by people or works autonomously within set limits and escalates when it hits them. Only 13% want fully autonomous AI. A quarter use no responsible-AI practices at all.
The product is the escalation path
Put the three together. Kalepa’s growth comes from making underwriters faster while they keep the sign-off. Capgemini finds banks deploying agents but almost none trusting them to run end to end. Rootstock finds manufacturers want agents that know when to stop and ask.
That’s the actual product being bought in 2026: a system that knows what it isn’t sure about and hands that part to a person, with a record of why. The vendors that win in regulated work won’t be the ones with the most autonomy. They’ll be the ones whose agents ask for help at the right moment. Kalepa’s cash flow suggests customers will pay for exactly that.