Teneo.ai Bet 25 Years on Zero Hallucinations

Most AI companies you have heard of this year are about three years old. Teneo.ai is about twenty-five. It was writing software to understand human sentences in 2001, back when a "chatbot" meant a scripted box on a corporate website and the transformer paper that started the current boom was still sixteen years away. That head start is the whole story here, and also the whole risk.

The company started life in Stockholm as Artificial Solutions, founded by Johan Ahlund, Johan Gustavsson and Michael Soderstrom. For most of its life it was a respected but low-profile natural language firm. In 2008 its chatbot, Elbot, won the Loebner Prize, a Turing-test-style contest, by briefly convincing human judges it was a person. Then, in August 2024, the company renamed itself after its own product and became Teneo.ai. The rebrand did something a press release rarely does: it changed how the market read the business overnight, from legacy NLP vendor to agentic AI platform. The engine underneath was already there.

The pitch today is narrow and specific. Teneo automates the phone and chat queues that run enterprise contact centers - the "Tier 1" support that eats budgets and patience in equal measure. And it makes a claim most vendors are too careful to put in print.

Zero hallucinations. No LLM wrapper offers this. Teneo.ai, on its own homepage

01 / The betControl the model, don't worship it

To understand why that sentence matters, you have to understand what almost everyone else built. The dominant pattern of the last few years is the "LLM wrapper": take a powerful language model, put a friendly interface on it, and let the model generate answers. It is fast to build and genuinely useful. It also, occasionally, makes things up - and a contact center that invents a refund policy or a warranty term is a liability, not a feature.

Teneo's answer is to invert the trust. Instead of letting the model speak freely, it puts a deterministic control layer in front of the model and orchestrates several models behind it. The language model handles fluency, the sound of a natural sentence. A rules-and-logic layer handles the facts, the actions, and the handling of personal data. The company describes it as using "as much or as little LLM as needed," with no lock-in to a single vendor's model. It is a less glamorous architecture than "just ask the AI," and it is quietly the direction a lot of serious enterprise AI is converging on.

The scale numbers are the part rivals notice. In early 2025 Teneo said it had passed 17,000 AI agents running in production. As CEO Per Ottosson put it, while much of the industry was "launching tens of agents with fanfare," the milestone was less a trophy than a signal about where the market was heading. Whether or not you buy every metric, the gap between "tens" and "thousands" is the kind of thing that is hard to fake in front of paying enterprise customers.

02 / The mathWhy a contact center would even care

The economics are blunt, and blunt is what sells to a CFO. Teneo puts the cost of an automated call at about $0.40, against roughly $5.60 for a live human agent. Multiply that gap across a large operation and the savings stop being a rounding error.

The reference deployment the company likes to cite is a global software provider running 84 million calls a year through Teneo 8, its voice platform launched in 2025. Teneo says that customer cut about $5.60 off each call and reported figures in the neighborhood of $264 million in annual savings, with total call resolution around 91% in any language. Other named customers span industries most people would consider hard to automate well: AT&T and Telefonica Germany in telecoms, Swisscom (which reported an 18-point NPS improvement), the medical-device maker Medtronic (which Teneo says freed roughly 36,000 agent hours), Circle K in retail and the insurer Folksam.

There is also a claim about speed that, if it holds, addresses the thing enterprises complain about most: time to value. Teneo advertises 10 days to a proof of concept, 30 days to a pilot, 60 days to production. Enterprise software is where pilots go to die in committee, so a vendor putting those numbers on the record is either confident or brave.

Major appliances sit at the center of the home, and when they fail, customers want immediate help that is accurate and actionable. Per Ottosson, CEO, on the January 2026 launch for appliance brands

03 / The catchGreat product, hard quarter

Here is the part the homepage does not lead with. Teneo.ai is a public company, listed on Nasdaq First North Growth Market in Stockholm under the ticker TENEO, after going public in 2019 through a reverse takeover of Indentive AB. Being public means the balance sheet is not a rumor - it is a filing. And the filings in 2026 are tense.

In mid-2026 the board disclosed a financing need and launched a strategic review, with the stated focus of selling all or parts of the business. It engaged J.P. Morgan and Redeye as financial advisors and the law firm Schjodt on the legal side. A legal dispute with a former US reseller partner had hit revenue, the company said, and its cash position was sufficient to fund operations for roughly one month at the current revenue level. It is also in talks with its lender over a SEK 250m loan and a minimum-liquidity covenant. The June 2026 annual meeting authorized the board to issue new shares, warrants or convertibles - the classic toolkit of a company that needs capital fast.

This is the honest tension of the story. The technology has real customers, real scale and a differentiated idea. The company holding it has, at the moment of writing, a runway measured in weeks rather than years. Both things are true at once, and pretending otherwise would be the kind of thing this platform exists to prevent.

04 / The fieldWho else is in the room

Teneo sits in a crowded market, and each rival represents a different theory of the same problem. LivePerson leans into "conversational commerce," guiding shoppers toward a purchase. Ada sells a no-code product built for support teams that would rather not involve developers. Rasa is the LLM-agnostic, developer-first option with strict logic for business-critical steps. Botpress is the open-source, build-it-your-way choice for technical teams. Microsoft brings the gravity of a platform giant. Against that lineup, Teneo's wedge is the combination that is genuinely hard to assemble: deep multi-turn understanding, many languages from one instance, and a deterministic layer that lets a regulated enterprise sleep at night.

What can you actually do with it? If you run a support operation, Teneo is a way to automate the repetitive front line - password resets, warranty checks, service scheduling, balance inquiries - across voice, chat, email and apps, in dozens of languages, while keeping a tight grip on what the AI is allowed to say and do. The promise is not a cleverer chatbot. It is fewer wrong answers at industrial scale, with the receipts to prove it.

As the industry is catching up and launching tens of agents with fanfare, our milestone isn't just a technological achievement - it's a turning point for the industry. Per Ottosson, CEO, on passing 17,000 agents

The Latin word teneo means "I hold" or "I grasp." For a company selling control and comprehension, it is a well-chosen name. The open question is whether the firm can hold on to itself long enough for the market it helped invent to reward the patience. Twenty-five years is a long time to be early. 2026 will decide whether it was too early or exactly on time.

05 / FAQQuestions people actually ask