The return-to-office argument runs on a shared assumption that both sides accept: that work happens when people are available at the same time, and the only question is whether that time is spent in a building or on a screen. Async-first organizations reject the assumption itself. They arrange work so that most of it does not require anyone to be available simultaneously, which makes the location debate largely beside the point. If the work is written down, handed off, and judged by its output, it matters little where or when the writing happened.
What async-first actually requires
This is not simply remote work with fewer meetings. It is a different operating discipline, and the companies that practice it are candid about its price. GitLab, which runs entirely distributed, builds on what it calls a handbook-first approach: a public, living document so extensive it would run past two thousand printed pages, because, in the company's own framing, asynchronous work is impossible without strong documentation. The logic is strict. If a colleague's question can block your work for hours or days while you wait for them to come online, the organization has not gone async; it has just moved synchronous dependency onto a screen. Documentation is what lets people serve themselves answers instead of waiting.
The conceptual backbone is a shift in what gets measured. Matt Mullenweg, describing five levels of distributed-work maturity, places the meaningful threshold at the point where work is evaluated by output rather than by hours or process. Below that line, a company merely recreates the office online, synchronous, meeting-heavy, presence-oriented, and gets remote work's dislocation without its benefits. Above it, the unit of account becomes the deliverable, and presence quietly stops being a management concern because it stops being a management signal.
Async-first, not async-absolute
The honest version of this model is not dogma, and its practitioners say so. GitLab's stated position is async-first, not async-only: it names specific triggers for reverting to real-time work, incidents, tight deadlines, live troubleshooting, brainstorming, and relationship-building, where synchronous time genuinely serves the work better. That nuance matters, because the failure mode of async is a team that refuses to get on a call when a call is obviously the right tool, and grinds through days of written back-and-forth to avoid it. Async-first is a default with exceptions, not an ideology.
It is also worth being clear about the limits of the evidence. The async case rests largely on practitioner accounts rather than randomized trials; Mullenweg's claim that a fully distributed organization outperforms any in-person one is an assertion without empirical backing, and should be read as advocacy, not proof. What the rigorous research supports is narrower and still useful: that autonomy and choice over working arrangements tend to raise both output and retention, which is the soil async-first grows in.
Measure the pump, not the presence
Underneath the tooling, async-first revives an old idea about what management is for. Semler describes the turn Semco made after he stopped counting hours: the company began asking whether the pump had been delivered rather than whether the worker had been present. That is the entire async proposition compressed into one question. If the deliverable is defined, handed off cleanly, and checked against a standard, the coordinates of the person who produced it are irrelevant. The organization that can ask and answer that question has no mandate fight to have, because it no longer runs on presence to begin with.
The catch is that async-first is harder to install than a mandate, not easier. A mandate requires a memo. Async-first requires the whole apparatus of clear ownership, written context, and honest output evaluation, the same capabilities whose absence pushes weaker organizations toward presence in the first place. That is the real trade. A company can fight the return-to-office war indefinitely, or it can do the harder work of making location stop mattering. The second path is not a shortcut. It is the price of never having to argue about the office again.
Sources
- GitLab (2026). All-Remote: Asynchronous work (handbook). Handbook-first practice: async work is impossible without strong documentation; async-first, not async-absolute, with named triggers for reverting to synchronous. https://handbook.gitlab.com/handbook/company/culture/all-remote/asynchronous/
- Mullenweg (2020). Five Levels of Autonomy. Level 4 defines distributed work by output rather than hours or process; the claim that distributed beats in-person is asserted without empirical evidence. https://ma.tt/2020/04/five-levels-of-autonomy/
- Bloom, Liang, Roberts & Ying (2015). Does Working from Home Work? (QJE). RCT: giving employees choice over arrangement raised productivity gains toward ~22% via selection and learning. https://academic.oup.com/qje/article-abstract/130/1/165/2337855