Typewise launched Nova today to handle the ongoing monitoring, testing, and tuning that customer service AI agents require, priced on resolved outcomes rather than seats.
Every company that deploys a customer service AI agent inherits a second, quieter job right behind it. Someone has to watch how the agent performs, catch the wrong answers before customers do, update its knowledge when a policy changes, and decide when it has earned the right to handle more on its own. Many teams either create a role for that work, hand it to whoever has spare time, or skip it and settle for a flatter, less accurate chatbot. Typewise, a customer service AI company based in Zurich and San Francisco, launched a product today aimed squarely at that gap.
The product is called Nova. Typewise describes it as an AI operator rather than a customer facing chatbot. According to SiliconANGLE's report on the launch, Nova builds, runs, and continuously improves the AI agents that handle a company's email, chat, and WhatsApp support, and does the maintenance work that would otherwise fall to a person. Typewise co-founder and chief executive David Eberle put the pitch directly: "Every company that deploys AI agents creates a new operations job. Someone must monitor them, update their knowledge, test changes and decide when they are ready for more responsibility. Nova is that 24/7 operations team."
The mechanics follow a supervisor and specialist pattern the company has been building toward for months. An earlier PR Newswire announcement from Typewise, published in February, described the underlying architecture: a supervisor agent reads a customer's request and routes it to a specialist agent built for that task, which pulls from connected systems such as a CRM, ERP, or billing platform to check an order, change a booking, or issue a refund. Companies set the guardrails in plain language rather than code, cap what an agent can do without approval, and get a logged record of every action. Nova's job is to keep that whole setup current: watching how agents perform, proposing fixes, and testing changes before they ship, according to Typewise's own site. That February release also described an early customer, Beurer, whose earlier pilot with a legacy vendor had stalled over high cost and integration friction; adopting Typewise let the company drop its reliance on outside integrators entirely.
Typewise points to two customers to back up the newer numbers. At HealGreen, a telemedicine platform in Hamburg, the company says its AI agents now handle 70% of incoming inquiries across email, chat, and WhatsApp, and fully resolve between 75% and 85% of those. At Beurer North America, Typewise reports the agents autonomously resolve 65% of incoming inquiries with a success rate above 90%, and that average handling time for the requests still routed to a person fell from 12 minutes to six.
For a business leader, the detail worth sitting with is the pricing structure more than any single statistic. Nova is priced on outcome rather than seat or license: a fully resolved request is billed in full, a partial hand-off to a human is billed at half, and an unresolved request costs nothing. That structure is confirmed on Typewise's page for smaller businesses and repeated in the SiliconANGLE report. It reframes the buying decision. Instead of paying for a platform and separately staffing or contracting someone to keep it accurate, the ongoing tuning work is folded into the per resolution price. Typewise has not published what that per resolution rate actually is, and knowing the pricing structure is not the same as knowing what a dedicated operations hire or an agency retainer would otherwise cost a given company, so the comparison here is directional rather than a figure a CFO could drop into a spreadsheet today.
That is also where the honest limits show up. The self-testing loop Nova describes, watching performance and proposing fixes, still requires someone on the customer's side to approve changes, review escalations, and decide where the guardrails sit. A one or two person support team with no documented policies to hand the agent will likely spend real time writing those down before Nova has much to work with. A company in a regulated space where every customer interaction needs a compliance sign-off before it ships will find the "test and adjust automatically" pitch less useful than the manual review it is meant to replace. And Typewise's own numbers, a gap of five to fifteen points between the share of inquiries an AI agent touches and the share it actually resolves on its own, are a reminder that a person is still doing meaningful work on the other side of even a well tuned deployment.
The interesting shift is not that an AI agent can answer a support ticket now. It is that a vendor is willing to sell the maintenance of that agent as a distinct, ongoing product, instead of leaving a customer to discover that job exists on their own after go live. Whether that operations work actually disappears, or simply moves from a line on someone's job description to a line on a subscription invoice, is the part worth watching over the next few renewal cycles.