What Remote Work Actually Did to Software Quality

For four years now, the argument over remote engineering has run on vibes. The maximalist case said distributed teams ship faster, cheaper, and happier. The reactionary case said code quality cratered, onboarding broke, and the data centers of software culture rotted out from the inside. By 2026, the data is finally good enough to answer.

We pulled from three sources: the DORA Accelerate State of DevOps reports for 2022 through 2025, the Stanford Institute for Economic Policy Research studies on remote productivity led by Nicholas Bloom, and the Microsoft Work Trend Index longitudinal series. We cross-referenced those with public engineering blogs and a non-public survey of engineering leaders we ran in early 2026. What follows is what the data actually says.

Throughput stayed flat. Quality moved a little.

The single most replicated finding across DORA, Bloom's SIEPR working papers, and Microsoft's index is that pure throughput — features shipped, story points completed, lines of code merged — did not change in any direction that survives a careful read of the data. Fully remote, hybrid, and fully co-located teams sit inside the same wide distribution.

What did change is the variance. Remote teams that managed the change well were on the upper end of the distribution. Remote teams that did not, were on the lower end. The gap between "managed well" and "managed badly" widened. The office, it turns out, was acting as a shock absorber for sloppy process. Remove it and process discipline shows up in the numbers.

The clearest signal of "managed badly" is code review depth. Microsoft and GitHub's joint research, published in their 2024 engineering blog series, found that fully remote teams without explicit review norms reduced their average comment count per pull request by roughly a third. Teams with explicit review norms — required reviewers per area, response-time SLOs, a "no rubber stamps" culture — held their depth steady or grew it slightly.

Onboarding got harder, then teams fixed it

The strongest case against remote work in 2022 was that junior engineers were being failed by it. That was true. It is less true now.

The fix was not technical. It was structural. Companies that hold onboarding satisfaction above pre-pandemic levels in 2026 have all done some version of the same three things: a multi-week structured ramp with a named buddy, frequent live pairing in the first quarter, and an explicit move-away from "ask in Slack and hope" as the primary support channel. None of these are new ideas. The shift was making them mandatory rather than aspirational.

Companies that did not make this shift continue to lose juniors at materially higher rates than they did pre-pandemic. The gap is large enough that it shows up in retention reports.

Incident rates: a modest improvement, with caveats

The most counterintuitive finding is on incident rates. Across the public postmortem record, severity-one incident frequency is flat to slightly down for engineering organizations that went fully remote, compared to their 2019 baselines. That is not what either side of the argument predicted.

The likely explanation is mundane: the move forced engineering leaders to write down processes — runbooks, oncall rotations, escalation criteria — that had previously lived as oral tradition. Writing them down made them better. The remote forcing function paid back in operational discipline.

The caveat is that the rate of complex incidents — the kind where the problem is not one component but the interaction between teams — appears unchanged or slightly worse, depending on how you slice the data. Cross-team coordination is the part of engineering that the office quietly subsidized, and we have not fully replaced it.

What the data does not say

It does not say that hybrid is universally better than fully remote. The hybrid premium in productivity studies is small and confounded by selection effects. It does not say that everyone should be in the office two days a week. And it does not say that remote work caused the broader 2023–2025 layoffs — those tracked interest rates and the AI-driven margin compression, not work arrangement.

The honest summary, four years in: managed well, distributed engineering produces software of comparable quality to co-located engineering. Managed badly, it does not. The differentiator is process discipline, not pajama policy.

Sources

• DORA Accelerate State of DevOps reports — dora.dev/research
• SIEPR working papers on remote work (Bloom et al.) — siepr.stanford.edu
• Microsoft Work Trend Index — microsoft.com/en-us/worklab/work-trend-index
• GitHub engineering blog series on review patterns — github.blog/engineering