Quentin Stafford-Fraser was working in the Trojan Room of Cambridge University’s old Computer Lab when he and a colleague rigged a grey-scale CCD camera, pointed it at the communal Krups coffee machine, and wired it through a video-capture card into a computer. The image was small and grainy. The coffee, by all accounts, was awful.
The problem was simple. The lab was spread across multiple floors. The coffee pot lived in one corner. Walking down to find an empty jug, then walking back up empty-handed, was the kind of small daily defeat that researchers in a multimedia networking group felt professionally obligated to solve.
So they solved it. And in doing so, accidentally invented the webcam.
The Trojan Room and the terrible coffee
Stafford-Fraser’s group had been working on multimedia over computer networks. Genuinely frontier work in 1991, the same year Tim Berners-Lee published the first website at CERN. They had leftover equipment from earlier experiments. A CCD camera. A video-capture card. And the long-standing tradition, going back to mainframe days when researchers booked overnight slots and lived on caffeine, that good computer science required hot coffee within arm’s reach.
Stafford-Fraser and his colleague wrote a server program that grabbed a frame from the camera every few seconds. Then they built a small X-Windows client that displayed a thumbnail-sized image in the corner of any researcher’s screen, anywhere in the lab. Glance at the corner. See the pot. Decide whether the walk was worth it.
Initially, that was the entire user base. A handful of computer scientists, in one building, checking on one jug.
Why grey, and why so small
The image was monochrome because colour CCD capture in 1991 was expensive and slow, and the team was reusing kit. It was small because that was enough resolution to tell whether the pot was full, half-full, empty, or, the worst outcome, present but containing only a brown crust at the bottom. The system ran on an ARM processor, the same architecture lineage that now sits inside almost every smartphone on Earth.
The Krups machine produced what Stafford-Fraser later described as truly terrible coffee, only palatable when fresh. That detail matters, because it is the entire engineering justification. A pot that tastes fine after sitting for an hour does not need a camera trained on it. A pot that turns to sludge in twenty minutes does.
The whole apparatus existed because someone, somewhere in the building, was always furious about a wasted journey.
1993: the pot goes public
The web happened. The HTML IMG tag had just arrived. People were starting to embed static pictures — institutional crests, diagrams, photographs of pets — into early Mosaic pages. Researchers in Cambridge wondered what would happen if the image a browser requested was different every time it asked.
Stafford-Fraser’s colleagues modified the existing server so it could also answer HTTP requests. In 1993, the XCoffee feed went live on the open web. Anyone with a browser, anywhere on the planet, could now check the level of a jug of bad coffee in a department in eastern England.
It became a popular phenomenon on the early web. Part of the appeal was the absurdity of using expensive research equipment for something so trivial. Part of it was something stranger. Up to that point, the web had been static. You clicked through hyperlinks and arrived at frozen pages. The coffee cam did something nobody had done online before: it dropped you, abruptly, into a real room in real time. Before this, watching a live image of a remote place required a broadcast truck, a satellite uplink, and a network with a programming schedule.
Now it required a URL.
The afternoon that became a category
The whole prototype took a short time to assemble, though Stafford-Fraser is careful to point out it was only possible because of years of prior work on multimedia network protocols in the group. That is how a lot of accidental inventions look up close. A quick hack sitting on top of a decade of slow, unglamorous research. A familiar pattern in the history of accidental discoveries, where the breakthrough arrives sideways while someone is trying to solve a much smaller problem.
By the late 1990s, webcams had escaped the lab entirely. Logitech, Creative Labs and Philips were shipping consumer USB cameras. Universities pointed cameras at fish tanks, ant farms, building sites, the Empire State Building, Mount Fuji, traffic intersections. JenniCam, Jennifer Ringley’s continuous broadcast of her dorm room, starting in 1996, turned the format into a cultural argument about privacy years before social media existed.
The global webcam market was valued at $6.69 billion in 2020, driven by security and surveillance. Every video doorbell, every Zoom call, every baby monitor, every nanny cam, every traffic-light enforcement camera, every Twitch stream traces back, in some way, to that grey square pointed at a Krups in Cambridge.
2001: fingers on the kill switch
The lab moved. The Computer Lab was relocating from its old building in central Cambridge to the new William Gates Building on the West Cambridge site. The coffee pot cam had served its purpose. In 2001, the team gathered around the server. The final image transmitted from the XCoffee feed showed the assembled researchers ending the stream together.
Its retirement made news. A camera nobody had asked for, pointed at a pot nobody really liked drinking from, had become important enough that its shutdown was an international story.
The pot itself was auctioned, with the proceeds paying for a much better coffee machine and a generous supply of beans.
What the grey square actually invented
The technical lineage is obvious. Server-pushed images, then MJPEG, then streaming video, then WebRTC. The lab had figured out, on a Tuesday afternoon, that the web could be live. That a URL did not have to return the same bytes twice. That a browser could be a window into a room.
The cultural lineage is stranger. The XCoffee cam was the first time the internet was used to solve a problem that was, in absolute terms, almost negligible. Was the pot full? By deploying disproportionate technical effort. That ratio, of small problem to large solution, has become a defining characteristic of the consumer internet. Smart fridges that text you when the milk is low. Apps that track which days your dog rolled in the grass. Doorbells that ring your phone in another country to show you a delivery driver leaving a package.
The coffee cam established the template: glance at a corner of a screen, see a faraway thing, decide whether to act. That same glance is now what hundreds of millions of people do every few minutes, all day, on phones. The Trojan Room set up the muscle memory.
The unintended cost of the glance
There is a less cheerful trace too. Stafford-Fraser has spoken about keeping multiple webcams pointed at his own garden, driveway and fields, a personal home-automation hobby. The number is striking only because in 1991 it would have been unthinkable. The same impulse that wired one camera to one pot now wires dozens of cameras to dozens of feeds in dozens of homes.
Workplaces have absorbed the same pattern at an industrial scale. Constant context-switching between platforms, notifications and redundant dashboards is chipping away at focus and mental health, a condition some have termed digital tool fatigue. The thumbnail in the corner of the screen multiplied. Then multiplied again.
None of which the Cambridge team could have predicted when they zip-tied a camera to a stand. They were trying not to walk down multiple flights of stairs for nothing. The brain treats small wasted journeys as real cognitive losses, which is why a hack that saves you twenty wasted trips a week feels, at the time, like genuine engineering.
What stayed in Cambridge
The Computer Lab is now the Department of Computer Science and Technology, in a building named after a man who had nothing to do with the original hack. The Trojan Room no longer exists in the form Stafford-Fraser knew it.
What persists is the gesture. Look at a small image in the corner of a screen. Decide whether the thing in the faraway room is in the state you want it in. Get up only if it is.
The numbers since 1991 are difficult to absorb. The global webcam market reached $6.69 billion by 2020. Roughly one billion CCTV cameras are now installed worldwide, with China accounting for over half of them. Ring alone has sold tens of millions of video doorbells in the United States, and police departments across the country have built formal request portals to access the footage. Zoom passed 300 million daily meeting participants during the pandemic, a figure that did not exist as a category before March 2020.
The original feed ran at roughly one frame every few seconds, at 128 by 128 pixels, in grey. The infrastructure that grew out of it now captures, by some industry estimates, more than two billion hours of video footage every single day. A jug of bad coffee in eastern England, watched by a handful of researchers, became the prototype for a planet that watches itself constantly.
