AI is pushing some workers back to the office—and others out of it
By
Imagine six engineers walking into a glass-walled room in a LinkedIn office building. One taps a badge. Technology in the room is linked to the project they're working on, and can identify where they left off the day before. An AI agent listening to the meeting can read and update the engineers’ project tracking system, and quickly surface earlier meeting transcripts as they debate and discuss their work. By the time they wrap, the agent has ordered their coffee, which is waiting for them downstairs in the cafe.
This is what an office designed for the AI era might actually look like, and LinkedIn is already piloting and planning the prototype.
Some CEOs are still arguing about days in office, pushing people back into open floor plans where voice-based AI tools can create noisy, distracting spaces as more people ditch keyboards for microphones. Other executives are doing the opposite, migrating back to remote-based work to allow for more seamless integration of AI by capturing conversations digitally. Still others think the office could be part of the solution. Aurelio David, who runs workplace experimentation at LinkedIn and led the pilot behind the new design, suggests AI has accelerated execution, but decision-making, alignment, and collaboration still lag.
The pandemic-driven shift to hybrid and remote work kick-started one of the biggest-ever changes in office design, as workplaces added more tools and more space for collaboration and connection. AI’s impact may be even bigger, but the direction of the change isn’t yet clear. Voice-based AI tools might lead to office designs that feel like cube farms from the past as workers seek more quiet and privacy. Teams might shift even more heavily to remote work to ensure all conversations and meetings land in an AI-accessible knowledge base.
Or, as LinkedIn is prototyping, the office might evolve to incorporate AI agents into physical workspaces. “The real opportunity is at the team level: speeding up how we brainstorm, assign tasks, and make decisions,” says Chris Rufo, a senior analyst on LinkedIn’s workplace design and build team. “The physical workplace needs to evolve to support that kind of rapid, collaborative iteration.”
A simple pilot that worked
David and Rufo’s team at LinkedIn has been working to find better solutions that combine workspaces and AI. The pilot design at LinkedIn, known as Pod Work Areas, are spaces for dedicated “pods” of six to 10 engineers, product managers, and designers. The areas consist of one meeting room, a few workstations outside of it, glass walls, and a key design choice: the rooms can’t be booked. The team owns the space, so there’s no searching for a spot to meet.
LinkedIn is also building agents that will be linked to the rooms’ built-in technology and primed for project management. The idea will be for employees to badge in, pull up a dashboard, ask what they worked on last, and quickly get the answer. Need a colleague who's working from home today? Tell the agent and they're dialed in on the side panel. GitHub and Claude will be integrated so the team can write, test, and commit code together in real time.
"The room takes over," says Rufo. "It's going to be more immersed into the way you're working." (To avoid capturing private conversations, the rooms are designed so the AI agent’s presence is opt-in. Team members have to elect to turn on both the recording and the voice interaction features in Microsoft Teams.)
An 11-week pilot last fall measured the designs against control teams in standard work areas. The pods produced 15% to 20% faster production cycles, measured by code shipped, project tasks completed, and other measures. Pod members said quick conversations were 43% easier, awareness of what others were working on improved by 27%, and brainstorming was 55% easier. People also came into the office more often, knowing the in-person patterns had purpose and support.
This July, LinkedIn will open a 120,000 square foot building in Sunnyvale, Calif., built around the design. It will feature 36 pods for engineers and a “library” for heads-down work, with about 90% of the building’s 500 people working in pods. The remaining 10% are senior leaders and support staff such as sales, finance, and HR teams that need a different layout.
The pull in the other direction
Not every team is being pulled into an office. Some are being pushed out, lured by the potential for putting the same technology to work.
David Purdy is the co-founder and CEO of Palladio AI, an early-stage startup. He and his co-founder live in the same town, and share a WeWork space three days a week. On Mondays and Fridays, they work from home, but it’s not just for flexibility or to save the commute. They want to talk to their AI tools out loud.
"You can't really do that in a normal way at an open office," Purdy says. Dictating to Claude or running a voice prompt through Wispr Flow in a public coworking space means popping into a phone booth or going home.
Purdy gave the broader problem a useful name: many-to-AI. Human-to-human collaboration works in shared space. Voice-driven human-to-AI work needs a phone booth or a closed door. A small team where each person is running multiple agents at once can produce a cacophony the office has to absorb.
After his conversation with Purdy, Stanford economist Nick Bloom predicted on LinkedIn that voice-based AI tools will drive higher rates of working from home, not lower. As CBRE chief strategy officer Annie Dean has said, “AI can’t hear the hallway conversation.” Even if coworkers are in a shared room and recording their meeting, AI may not be able to distinguish voices, making it harder to attribute who said what and assign action items.
Why both can be true
The two stories appear to contradict each other. They don't. They describe different kinds of work, and the data supports both.
Tech industry leasing rose to 22.7% of all new office activity in Q1 2026 in the US, up from 15.3% a year earlier, according to CBRE. AI companies account for 8.1 million square feet of tenants actively seeking space across the five largest US tech cities. Even the most AI-forward companies want to bring people together to build connections.
Coworking is growing just as fast. Industrious, the flexible workspace operator CBRE owns, signed 52 locations in 2025, up 58% year over year, with the company projecting 100% unit growth in 2026. Many companies want the flexibility to bring people together, but also get heads-down work done remotely with AI.
Dean has a theory about why both are true at once. "There will be many more people building their own companies," she told me. "If you're working with a team of agents and one other person…a community like a coworking space [offers] that sense of continuity and community."
In a world swimming in AI agents, some combination of connection-focused workplaces, coworking facilities, and more frequent team gatherings might be essential to preserve connection, engagement, and trust among teams. Recent studies from Anthropic, Cisco, and Upwork all point to heavy AI users losing a sense of connection to teammates.
Dean sees the value of community growing even more. She has worked remotely for years, but now finds that working with AI is different. "I never felt lonely or kind of disconnected as a remote worker," she said. "But when I take on projects that are very AI-based, I find myself totally fatigued from just speaking to a machine all the time."
What this means for leaders
As workers interact with AI agents more and teammates less, here are steps leaders can take today:
Give your workplace team a seat at the table. Your AI steering committee or “chief work officer” already includes HR and IT, but the workplace team is likely still treated as a cost center. Add those leaders to goal-setting, planning and experimentation.
Design for the acoustic problem coming your way. Voice dictation and AI agents plugged into meetings will make open floor plans feel like call centers. Start mapping spaces into behavior-based zones: quiet library spaces, raucous cafes, and collaboration zones.
Run a pilot before launching a policy. LinkedIn’s 11-week test delivered great results, but even if it didn’t, David’s team would have learned what wasn’t working before investing in redesign, or increasing return-to-office demands.
What else to read:
- The companies making their offices more ‘fiercely human’ for the age of AI
- A framework for designing offices for better collaboration
- Why we can’t stop talking about remote work nearly six years later
