Industries / Construction and property / Real estate / Coworking
Booking desks and rooms
This is the software and the marketplaces through which people reserve a desk, a meeting room or a day pass, and the systems operators use to run bookings, member records and invoices.
6
Milestones achieved, from 2010 to 2025.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2010
Operator software goes on sale
Cobot began in 2009 as the in-house tool of co.up, a coworking space in Berlin, and its makers began offering it to other spaces in 2010. Software that runs a space's bookings, members and invoices became a product an operator could buy. Nexudus, founded in 2012, followed.
Credit
Alex Lang, Kristina Schneider and Thilo Utke (Cobot); Adrian Palacios and Carlos Almansa (Nexudus)
2011
Booking an office by phone
LiquidSpace launched an app that showed nearby unused offices and lounges that people could reserve, which TechCrunch called "an Airbnb for workplaces", and tested it with pop-up workspaces in Austin. In 2015 it opened an API, a published way for other software to exchange data with it, so that operators' systems and calendars could send it live availability.
Credit
Mark Gilbreath and Doug Marinaro, LiquidSpace
2018
Calendars release empty rooms
In August 2018 Google Calendar began freeing a booked room automatically when all but one guest had declined. Microsoft's room panels and Exchange Online now cancel a booking when nobody checks in soon after the start, with a default wait of 10 minutes. Releasing rooms after a no-show became a built-in setting in the two main office calendars.
workspaceupdates.googleblog.comlearn.microsoft.comlearn.microsoft.com
Credit
Google; Microsoft
2020
Hourly rooms on long leases
Breather opened in New York in 2013, renting private rooms by the hour at around $20 an hour that members opened with its app. It grew to more than 400 rooms it leased itself. In December 2020 it laid off most of its staff and its US and UK companies entered insolvency to hand back 315 US and 40 UK spaces. Its chief executive said Breather "in its current form as an operator, doesn't make sense."
LessonShort bookings stopped when offices emptied, and the long leases behind them did not.
Credit
Breather (a failure)
2020
Hours at the largest operator
WeWork began On Demand in August 2020 at 12 sites in Manhattan and Brooklyn, letting anyone book a desk for $29 a day or a meeting room from $10 an hour in an app, with no membership. By November 2020 it covered about 160 buildings in 11 US markets, by WeWork's count.
Credit
WeWork
2025
One owner, many systems
Yardi Systems makes Yardi Kube, software that operators use to run their spaces, and runs the CoworkingCafe marketplace. Its affiliate Cupar Grimmond was set to take about 60% of WeWork in the 2024 bankruptcy exit, and in January 2025 Yardi bought two more booking marketplaces, Deskpass in North America and Hubble in the UK. Operator software, three marketplaces and the largest coworking operator came under one owner, joined by private links, not a public standard.
Credit
Yardi Systems
Watch for
An AI front desk
Nexudus says its AI Assistant helps members "create, view, modify and cancel bookings" by chat, email, WhatsApp or voice. Each booking goes through a draft, the member's confirmation and then the booking itself, and a payment link where one is due.
Achieved whenAn operator or reviewer outside Nexudus publishes how many bookings the assistant made over a set period and how many staff had to correct.
Claimed by
Nexudus, June 2026
General assistants that book
The Model Context Protocol (MCP) is an open way for assistants such as Claude or ChatGPT to use another company's software. Cobot says administrators can "book rooms, check availability and get reports" from such assistants through its MCP connection, and Spacebring says operators can work with their data in Claude and ChatGPT through its MCP server. Each connection reaches one software system only.
Achieved whenAn outside party publishes a test of bookings made through these connections, with success and error counts.
Claimed by
Spacebring, October 2026; Cobot, help page read in November 2026
Other operators in one app
WeWork says its app can book about 2,000 locations, including spaces run by more than 900 other operators in 21 countries, open only to companies that subscribe to its WeWork Workplace software.
Achieved whenAn outside party counts the partner listings that can actually be booked and confirms that bookings there are confirmed at once.
Claimed by
WeWork, February 2026
Open
01
An open availability feed
An open schema, a shared description of the data, for spaces, bookable items (desk, room, day pass), time slots and booking requests, with a reference feed and working adapters that turn two or three management systems' own interfaces into the same feed.
The briefCredit · unclaimed02
Rooms booked but unused
An open, anonymised dataset that pairs each meeting-room booking with whether anyone came, across several coworking spaces, and a test of no-show release rules against it.
The briefCredit · unclaimed03
One request, many operators
An AI agent that takes one request, such as "a room for four near the station on Tuesday from 10 to 12", searches and books across several operators' booking pages or connections, and logs every attempt and why each failed. Testing uses free resources set up in trial accounts, or stops at the payment page, and cancels each test booking at once.
The briefCredit · unclaimed
Open 01 · the brief
An open availability feed
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open schema, a shared description of the data, for spaces, bookable items (desk, room, day pass), time slots and booking requests, with a reference feed and working adapters that turn two or three management systems' own interfaces into the same feed.
Why it is still open
Every link found is one company's own interface: LiquidSpace's, Upflex's, Cobot's and Nexudus's. In a 2026 survey of 334 operators by OfficeRnD, a software seller, 68% named poor links between systems as their biggest technology problem, and operators had on average 1.65 of five core systems connected. The UK sport and fitness sector has an open data standard for this kind of job, OpenActive, whose 2017 data model left "availability, booking process and membership packages" out of scope and handed them to a separate booking specification. Nothing like it was found for workspace.
Built with what exists
OpenActive's data model and its RPDE feed format, a standard way to publish a stream of changes that other apps poll; schema.org's MeetingRoom and Reservation types; JSON Schema and OpenAPI 3.1; trial accounts on Cobot, Nexudus and OfficeRnD and their interfaces.
First step
Open trial accounts on two systems and map each one's resource and booking fields onto OpenActive's facility and slot fields in a one-page table, posted for comment.
Done looks like
A public repository with the schema, a validator and adapters for at least two systems. One client lists availability on each and creates and cancels a free test booking on each, with a published log of 50 create-and-cancel cycles per system, the success count and the median delay before a change shows in the feed.
Open 02 · the brief
Rooms booked but unused
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open, anonymised dataset that pairs each meeting-room booking with whether anyone came, across several coworking spaces, and a test of no-show release rules against it.
Why it is still open
Every figure found comes from a seller of sensors or software, each on its own definition, and they disagree. VergeSense says 11.3% of meetings are "ghosted", booked with nobody using the room, from its customers' data. Other vendors give 25% or more. None of the data can be checked, and none of it is from coworking spaces.
Built with what exists
Booking exports from operator software or from the Google Calendar and Microsoft Graph interfaces; check-in logs from room panels; simple motion or door counters that record presence without identifying anyone; Python and pandas; Zenodo to give the dataset a permanent citation.
First step
Agree with one operator to export a month of room bookings, put a counter in two rooms, and write down the definition before collecting.
Done looks like
At least 5 spaces, 40 rooms and 8 weeks, with the definition of an unused booking fixed before collection. Someone else rerunning the published notebook gets the same rate. A table shows, for release waits of 5, 10, 15 and 20 minutes, the share of unused bookings recovered and the share of real meetings released by mistake.
Open 03 · the brief
One request, many operators
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An AI agent that takes one request, such as "a room for four near the station on Tuesday from 10 to 12", searches and books across several operators' booking pages or connections, and logs every attempt and why each failed. Testing uses free resources set up in trial accounts, or stops at the payment page, and cancels each test booking at once.
Why it is still open
Every AI booking tool found works inside one system, such as the Nexudus assistant and the Cobot and Spacebring connections. The one connection found that spans operators, Workin.space's, is "completely read-only", and booking still happens on its website. No published tally of an agent booking across operators was found.
Built with what exists
Browser Use 1.0 for booking pages with no interface (see the Note "Deleting 80% to start over"); MCP connections where they exist; a small model such as Muse Ember to read slots and route steps (see the Note "The jobs agents repeat"); Playwright to replay each run. Include runs where several agents chase the same slot: Anthropic Fellows simulations found that one agent per person over a shared resource did worse than one agent serving the group (see the Note "Many agents, one budget").
First step
Set up free test rooms on one or two trial systems, write 20 test requests and do them by hand first to get a human baseline.
Done looks like
A public log of at least 100 attempts across at least 5 operators on at least 2 software systems, each with the request, steps, result, time and a screenshot, a success rate per operator, and no real charges.