It builds robots whose entire job is to watch your TV - so the bug never reaches the couch.
It is 3 a.m. in a lab outside Paris, and a small black box is changing channels on a real set-top box. Not a simulation. A real box, a real remote signal, a real screen. The machine waits for the picture, looks at it the way a person would, and decides whether the video is good. Then it does it again. And again. On a phone in Denver. On a smart TV in Singapore. On a streaming stick in a rack in New York.
This is Witbe. The company makes robots and software that test and monitor video services the way actual viewers experience them - continuously, on real devices, across any network. Its customers are the operators and broadcasters whose names you already pay every month. Witbe's machines catch the frozen frame, the missing channel, the ad that never loads, usually before a single customer picks up the phone to complain.
The real measure of quality is what customers see and experience.- Witbe's founding belief, unchanged since 2000
In the early days of the French internet, Witbe's founders kept hitting the same wall. Customers reported problems their engineers simply could not find. Every gauge on the backend read normal. The packets flowed. The servers hummed. And yet the experience, at the other end of the wire, was broken.
That gap has a name now: the difference between Quality of Service - is the infrastructure working? - and Quality of Experience - does it actually look right on the screen in your living room? Most monitoring tools of the era measured the first and quietly assumed the second. Witbe decided the assumption was the whole problem.
Video made the gap worse. A stream can be technically "delivered" and still stutter, freeze, mislabel a channel, or drop an ad. Multiply that across set-top boxes, smart TVs, phones, tablets, and a dozen streaming apps, each with its own quirks, and the only honest way to know if it works is to look. On the real device. Like a human.
Customers reported issues the engineers couldn't detect at the source. The fix wasn't a better gauge. It was a different vantage point.- The origin problem, retold
In 2000, fresh off helping build the French internet with their company Oleane, Jean-Michel Planche and Marie-Veronique Lacaze started over. They asked one question - "Who is the best at delivering quality customers can actually see?" - and named the company after it. Witbe. W-I-T-B-E. Who Is The BEst.
The bet was unfashionable. Everyone else was instrumenting the network. Witbe proposed the opposite: build a robot that behaves like a real user, sit it in front of a real device, and let it judge quality from the couch's point of view. It sounds obvious in hindsight. Most good bets do.
President & CEO. Runs the company that turned "look at the screen" into a public, profitable business.
SVP Strategy, Innovation & Technology. Internet pioneer; the person who kept asking why the gauges lied.
At the center is the Witbox - a hardware robot that physically controls a real consumer device and captures what comes out of it. It is "non-intrusive" in the most literal sense: it never logs into your servers or reads your backend. It just watches, the way a customer with very good eyesight and infinite patience would. Around it sits a software suite that turns watching into data.
The Remote Eye Controller lets a QA engineer in one city drive a device sitting in another. Smartgate runs test scenarios around the clock and fires real-time alerts the moment a stream breaks - a virtual NOC for video. And quality itself gets graded by algorithms with names only an engineer could love: VQ-MOS and VQ-ID, scoring picture and audio the way a trained human eye would.
Robots that control real STBs, smart TVs, phones & streaming devices - in the lab, the field, or the cloud.
Drive any connected real device from anywhere. QA teams stop flying to the hardware.
24/7 test scenarios, incident detection, and real-time SmartGate alerts to ops.
Non-intrusive testing: never touch the backend, just watch the screen. It turns out that's the only place the customer actually lives.- How the Witbox works, in one line
Then, in 2025, Witbe added a brain. Witbe Agentic AI lets teams describe a test in plain English - "open the app, play a 4K title, skip the ad, check the picture" - and have it run across many devices at once. It is a Gen-AI-first, human-in-the-loop system; the robot proposes, a person stays in the loop, and the old deterministic scripts still run underneath where certainty matters. The Witbox, after 25 years, learned to read instructions.
Jean-Michel Planche and Marie-Veronique Lacaze leave the internet-pioneer life at Oleane and bet on Quality of Experience.
Witbe builds the Witbox: continuous, non-intrusive testing on real devices, across real networks.
Offices open in New York, San Jose, Denver, Montreal, London and Singapore. Hardware gets assembled on Long Island.
Revenue hits EUR 26.1M, up 36% - proof that "look at the screen" is a real, growing business.
Natural-language test creation arrives: describe the test, run it on any device, keep a human in the loop.
Witbe debuts Agentic AI in Amsterdam and wins two CSI Awards, one with partner TAG Video Systems.
The customer list reads like your monthly bills: Comcast, Verizon, Cox, Peacock, Orange. These are companies that can build almost anything in-house and choose, instead, to let Witbe's robots do the watching. That is its own kind of proof.
When Comcast, Verizon and Orange outsource the looking, "look at the screen" stops being a slogan and starts being infrastructure.- Why the customer list matters
Witbe also stopped trying to see the whole chain alone. Its partnership with TAG Video Systems stitches Witbe's real-device, on-the-screen view together with TAG's real-time content matching - source to screen, end to end. The pairing won a CSI Award. Hardware credibility helps too: the Witbox+ was named Best in Market by the Media & Entertainment Awards and flagged by BTR's Diamond Technology Reviews.
Strip away the robots and the algorithms and Witbe is a single, stubborn idea: the only verdict that counts is the one from the couch. Everything else - the dashboards, the green lights, the packets per second - is a proxy that occasionally lies. Witbe spent a quarter-century building machines to stop trusting proxies.
That stubbornness scaled into something rare: a profitable, publicly traded company in a niche most people never knew existed, doing one thing with unusual conviction. Not the biggest software company on Euronext. Possibly the one most committed to a single sentence.
Every year there are more devices, more apps, more codecs, more ad-insertion tricks, more ways for a perfectly healthy network to deliver a broken picture. The job of looking - really looking, on the actual screen - only gets bigger. Agentic AI is Witbe's answer to scale: if a human can't watch every stream on every device, teach a machine to watch like a human and then point it everywhere at once.
So return to that lab outside Paris. The black box is still changing channels at 3 a.m. But now you can tell it what to check in plain English, it runs the same instruction across a wall of real devices, and it grades each picture the way you would. The robot got smarter. The conviction didn't change. Someone, somewhere, is about to watch a stream that works - and never know how close it came to not.
The picture either looks right or it doesn't. Witbe's whole company is the refusal to guess which.- The closing argument