Open Offices Were Already Failing. AI Just Made It Obvious.
· 4 min read · Nic Keating
The open-plan office was a poor fit for deep knowledge work long before AI. Now that so much work happens through a microphone — dictation, voice prompts, live transcripts — babble noise has turned an old irritation into a measurable tool-quality problem.
For more than 150 years, office work has been shaped by a mechanical inheritance. The QWERTY keyboard was designed in the 1870s to stop typewriter levers from jamming, then carried forward into the computer age so seamlessly that a typist from the 19th century would still recognise much of how we work today. That interface quietly shaped everything around it: rows of desks, heads down, screens forward, and a default assumption that productive work looked like people sitting still and typing.
In the mid-20th century, designers tried to improve on that model. Post-war Germany produced Bürolandschaft — "office landscape" — a genuinely humanist idea that offices should be organised around real workflows and human relationships, not rigid hierarchy. In the United States, Robert Propst's Action Office for Herman Miller followed similar principles: flexible, varied spaces designed for knowledge workers, not factory hands.
The open office did not begin as a cost-cutting exercise. It began as a reform.
But the ideal did not survive contact with corporate incentives. What organisations adopted at scale was not the nuanced philosophy. It was the cheaper, denser, easier-to-measure version: more people per floor, fewer walls, maximum visibility, and a persistent belief that proximity would automatically produce collaboration.
That promise was always shakier than it sounded. Workers complained for decades about noise, constant interruption, and the difficulty of sustained concentration in environments designed around constant exposure. Research gradually caught up with the intuition. A Harvard Business School study by Ethan Bernstein and Stephan Turban, using wearable sensor data to track actual behaviour across a Fortune 500 company, found that transitioning to an open-plan layout reduced face-to-face interaction rather than increasing it, with staff retreating to email and messaging instead.
And yet the model endured. Because its main benefit was easy to count, and its main costs were not. Lower space costs and higher desk density were shown in a spreadsheet. The cumulative cost of distraction, withdrawal, and fractured attention was not.
The open office was already a poor fit for deep knowledge work before AI arrived. What has changed is that a growing portion of knowledge work now involves microphones: dictating notes, prompting AI tools aloud, generating meeting transcripts, and thinking through problems in real-time conversations with digital assistants.
That addition changes the stakes in a specific and measurable way. A noisy office is not just distracting to people; it is also hostile to systems that depend on clean audio. Research on speech recognition performance shows that babble noise — the overlapping human voices that dominate open floors — is the most damaging noise type for AI transcription, because it occupies the same frequency range as the target speaker. In clean environments, modern speech recognition tools achieve very high accuracy; drop them onto an open floor and that accuracy falls away just when people are starting to rely on it.
There is a second concern that older debates about open offices only partly anticipated. In an open-plan environment, a stray conversation was once merely overheard. Now, depending on how AI tools are configured and what data policies are in place, it can also be captured, transcribed, stored, and searched. That is less a legal inevitability than a design and governance problem, one most organisations have not yet seriously addressed.
So the case against the open office is not that it suddenly stopped working. It is that it never worked as well as promised, and AI makes that failure operationally measurable for the first time. What was once written off as employee irritation is becoming a tool-quality problem.
The answer is not to abolish offices or retreat to remote work. It is time to stop pretending that every kind of work belongs in the same acoustic space. Focused individual work, voice-based AI interaction, and genuine face-to-face collaboration are distinct modes, and they have been awkwardly blended for decades.
Separating them is not a radical redesign. It is a correction that was overdue before AI arrived. In practice, it means quieter booths for individual focus and AI interaction, enclosed spaces for sensitive conversations, clearer zoning between concentration and collaboration areas, better microphones for teams actively using voice tools, and thoughtful policies on what gets recorded and when.
The office we inherited was designed around a largely silent, screen-based workforce. Adding voice as a serious work interface does not require tearing down the building. It does require being honest about a design assumption we borrowed from the typewriter age and never quite questioned.