Napster’s next act is teaching AI agents to be good coworkers
The potential customer gets straight to the point. “Why should I care about this call today?” she asks. The salesperson struggles to explain that every worker could have their own team of AI agents. But she does not let him finish. “How is that different from the tools we already have?” A message appears on his screen from Sami, his coach, who is listening to the call: Ask her about her priorities first, pitch second. He tries again, but the customer asks him for a clear number. “Honestly, saving me two hours a week sounds pretty modest,” she says. Luckily for him, this customer is not real. She was created using information from her public LinkedIn posts so he could practice in private. He needs the practice: Sami gives him a score of 42 out of 100 and kindly tells him to try again.
This may seem harsh, especially because Sami is not a human. She is one of seven AI agents in a sales team that Napster will launch this fall. (Yes, that Napster. More about that in a moment.) Unlike the many coding agents that developers create by the hundreds and then quickly throw away, Napster's agents are designed to work as long-term coworkers. They have names, faces, memories that last, and their own personalities. They aim to learn how work gets done inside the companies that use them. Napster is starting with sales because the average salesperson spends less than half their time actually selling. Much of the rest of the week goes to following up with people and coordinating tasks. If AI agents can take on a good share of that work, they can give salespeople more time to sell and quickly become worth their cost.
“We want to get to a point where, by the end of next year, every employee in every company has a dedicated crew of AI agents no matter what they do,” says Ziv Navoth, Napster's chief product officer. “Every executive has a chief of staff and a team behind them. Most employees never will. We want to give everyone that.”
For older millennials who may be wondering what happened to Napster: The company changed owners many times over the years. Last year, Infinite Reality bought it, along with Touchcast, a company known for its work on AI agents. All three companies then took the name Napster because it was much better known than the others. “Twenty-seven years ago, we democratized access to music,” Navoth adds. “Today, we’re democratizing access to expertise.”
WHAT HAPPENS AT WORK STAYS AT WORK
Corporate America recently had a short-lived craze for using as many AI tokens as possible, sometimes called “tokenmaxxing.” It ended with high costs and stricter rules for how employees could use AI. Napster believes the same kind of problem could happen as companies start using AI agents. “It’s agents gone wild,” Navoth says sadly. “We don’t know who’s building them, where they are, and what they have access to.” Agents that run without proper control can also create security problems, spread company data where it should not go, and perhaps even end the world. Napster's answer is to give workers a standard group of agents, with each agent responsible for a different area of work.
For example, the sales team starts with Jane, a chief of staff who checks your email inbox and calendar every morning. She puts together a list of the most important tasks and prepares draft email replies. Aaliyah researches potential customers and creates reports about them. Diego keeps HubSpot and other tools up to date so the salesperson does not have to. Unlike ordinary business software, these agents learn how their human coworkers like to work instead of making people follow new routines. Just as importantly, what the agents learn stays inside the company and becomes something the company owns. It does not become part of the same general AI assistants that competitors also use. “It’s great that everyone gets better,” Navoth says. “But your differentiation just disappears.”
BUILDING A “COMPANY BRAIN”
Napster calls its approach a “large persona model.” It is a secure memory system that keeps information over time and works on top of whichever large language model powers the agent. At the beginning, persona models give each agent a set of standard instructions for how to do its job. Customers can change these instructions to match their company's rules and usual ways of working. For example, Napster's agents will not approve unusual payment terms by default. But they can quickly learn to write requests asking for an exception instead of simply saying no every time. “That IP is what separates you from a generic offering, and it stays with the company,” Navoth says. Some people worry that AI agents will lead company leaders to cut large numbers of jobs. Napster takes the opposite view: AI should help workers become better at their jobs. It believes that is the job of AI.
Napster's agents are designed to know only what the company tells them. But company knowledge is often not neatly stored in a customer management system. Instead, it is spread across Google Drive, SharePoint, PDF files saved on computers, and even inside employees' heads. This kind of information, known as “unstructured knowledge,” has been a problem for business software for a very long time. Napster addresses this with what it calls a “company brain.” This acts as a trusted source of information that its agents can search. If an agent cannot solve a problem itself, it directs its human coworker to the right information at the right time.
The same idea is behind Napster Pulse. It uses AI-led interviews to help managers and company leaders learn more about each team member. They can find out what people are working on, what problems are slowing them down, and even how they feel about their jobs. “A survey tells you what question people ticked. A conversation tells you what they meant,” Navoth says. “Until now, nobody could have that conversation with a thousand people at once.”
Pulse started as a tool Napster used inside its own company. It will launch as a public beta, meaning an early version that people can try, as an addition to the business plan for the Napster app. “Our CTO wanted to hear from every engineer, every week,” he adds. “It doesn’t replace the conversation with your manager. It tells your manager what the conversation should be about.”
As Napster's sales agents learn more, they will take on more of the mental work that fills people's days. This should give human workers more time to do the things that only people can do. Practicing sales calls is not one of those things. “In most situations, you don’t have a colleague with time to practice, so you just don’t,” Navoth says. That is about to change. For now, Sami coaches salespeople during practice sessions. Soon, she will be able to give them advice directly while they are on a real call, giving them time to improve their score and make the sale. “Just wait until it’s in real time,” Navoth promises. “This stuff is like science fiction.”
