How AI Agents Will Disrupt Work — Starting With This Job

AI can already summarize emails and schedule meetings to make work easier. AI agents, however, go further by handling tasks on their own without needing you to step in. They act like junior teammates, not like simple assistants, and can follow instructions, adapt to feedback and improve over time. 

“We are in the early days of developing AI agents,” Jai Das, cofounder and president of Sapphire Ventures, a venture capital firm with over $11 billion in assets, said on Monday at the AI Agent Conference in New York City. That early-stage status means most of the truly powerful systems are still experimental. The basic capabilities are here, but best practices are still being worked out in real time.

Das caveated that AI agents currently require a human in the loop to check what they are doing. Users are now building agents that write code, check the code and run it. He predicted that over time, the amount of work that people do with agents will increase as the technology proves itself and builds trust. 

“Agents are about doing the tasks and working as a person,” Das said, pointing to customer service and coding as areas where AI agents excel. “One of the things we observed is how good the agents and models are going to get.”

According to Das, one job where AI agents are rapidly improving and could take over tasks is financial analysis. Agents are performing tasks like compiling data and creating financial models. They can pull in information from multiple sources, clean and structure it, then generate forecasts or scenario analyses that would normally take an analyst hours or days.

“Companies will have to adapt,” Das said. 

One big challenge that companies are facing with AI agents is the cost of running them. 

“Cost is going to be a really big issue,” Das said. “Agents are going to be running for hours, if not days.” Because AI agents often operate continuously in the background, the computing bills can add up quickly.

AI-native startups

Das highlighted a “bifurcation” among startups. Some startups have launched in the past few years and are AI-native, or have been built from day one with AI at the core of their products, workflows and decision-making, rather than adding AI later as a feature. Then there are older startups that have more human engineers on staff and are trying to catch up. 

The key to becoming AI-native is “a focused CEO,” Das said. He noted that AI-native companies have different business models. They are selling outcome or usage-based pricing, not subscription models. Instead of charging a flat monthly fee for software access, they might bill customers based on the number of tasks the agent completes. 

“I don’t think everybody will make it,” Das said. “That’s just the way technology works. I think the companies that will make it are the ones that take a radical approach.” In his view, the winners will be those willing to redesign their organizations around AI agents.