Move Fast and Break Things Is the Longwall Method for Software

The phrase started as a Facebook internal slogan, painted on walls in the early Menlo Park office. It meant something narrow and operational: ship code quickly, fix bugs in production, do not let perfectionism slow you down. Then it became something else. It became a philosophy. Move fast. Break things. Optimize for speed. Treat whatever falls apart as somebody else's problem, or as a feature, or as the cost of progress. The second thing I thought when I read about it was: that sounds exactly like the longwall method. The first thing was: I wonder how many people realize the longwall method failed.

Trist and Bamforth (1951) did not set out to study technology. They set out to study why productivity had collapsed in British coal mines after mechanization. The mines had introduced the longwall method, a technical reorganization that split a single shift of miners into three separate specialized shifts. Before the longwall method, miners worked in small autonomous groups that controlled the full extraction cycle. They knew each other, they made joint decisions in real time, they had a shared identity. The longwall method optimized for continuous production. Each shift did one piece of the work and handed the result to the next shift. On paper, the technical subsystem was more efficient.

What actually happened: absenteeism rose. Conflict between shifts spiked. Miners on the second shift blamed miners on the first shift for problems they inherited. Miners on the third shift blamed the second. The social fabric that had held the system together, the mutual obligation, the informal coordination, the collective responsibility for output, was gone. And productivity fell below pre-mechanization levels. The technical optimization destroyed the social system, and the social collapse dragged technical performance down with it. The mine owners thought they were modernizing. They were actually dismantling the only thing that made the old system work.

This is the part I cannot stop thinking about. The longwall method looked rational. It divided labor, it reduced idle time, it increased throughput on paper. Every single assumption was reasonable taken in isolation. And every single assumption was wrong because it assumed the social subsystem was a nice-to-have, not a load-bearing structure.

When I read about Facebook optimizing its newsfeed algorithm for engagement and watching democratic discourse fracture in Myanmar and in the 2016 U.S. election, I see the same assumption. The technical subsystem was optimized: viral content got distribution, time-on-platform went up, ad revenue grew. The social subsystem, community trust, information integrity, the shared factual ground that makes democratic deliberation possible, was treated as an externality. Cambridge Analytica was not a bug in the architecture. It was the architecture working exactly as designed, just with consequences the design never accounted for. Facebook did not mean to destabilize elections any more than the mine owners meant to destroy miner solidarity. They just did not think the social subsystem was something that needed its own optimization target.

Uber is the same pattern under different lighting. The early Uber culture, immortalized by Susan Fowler in 2017, optimized for growth, market capture, and engineering velocity. The technical subsystem was brilliant: real-time matching, dynamic pricing, seamless payment. The social subsystem, workplace dignity, gender equity, driver livelihood, the norms that make a company habitable for the people inside it, was treated as friction. Move fast and break things, applied to an organization, means optimize the metrics, ignore the social costs, and call the people who complain slow. The growth numbers were real for a while. The culture collapse, when it came, was expensive.

Theranos took the same logic further into territory that should have been obviously dangerous. The technical subsystem was optimized for narrative: valuation, media coverage, the appearance of breakthrough innovation. The social subsystem was the scientific community, peer review, regulatory oversight, the norms of validation that make medical technology trustworthy. Those norms were not friction. They were the infrastructure that makes a blood test mean something. Theranos treated them as obstacles to moving fast, and the company turned into a fraud. I think the interesting thing is not that Holmes lied. It is that the "move fast" culture provided cover for treating scientific validation as optional. The philosophy itself became the justification for bypassing the social system.

Twitter's post-acquisition restructuring made the pattern visible in real time. The technical subsystem was optimized for cost reduction: headcount cuts, infrastructure consolidation, feature velocity. The social subsystem, advertiser confidence, user trust, community norms, the unwritten rules that keep a platform from descending into chaos, was not factored into the optimization function. Advertisers fled. Users left. The platform did not break in a technical sense. It broke in the sense that the people who made it valuable stopped believing in it. Sarker et al. (2019) would call this a Type I mistake: treating the technical and social as independent when they are genuinely interactive. Only about 13% of IS research operates at genuine sociotechnical interaction, what Sarker et al. call Type IV. The rest treats technology as background context or treats the social as something that follows from the technical. I think "move fast and break things" is what happens when an entire industry normalizes operating at Type I.

I wrote about the original sociotechnical systems theory recently, about Trist and Emery's concept of joint optimization and what it actually costs. The core claim is not complicated. Any organization has a technical subsystem and a social subsystem. Optimizing either one alone degrades the performance of the whole. Joint optimization means designing both together, explicitly, with the understanding that changes to one ripple through the other. The theory does not say be nice to users. It says the subsystems are interdependent, and treating them as independently optimizable is actively harmful.

The thing that gets me is that we have had this theory since 1951. Trist and Bamforth published their findings seventy-five years ago. Bostrom and Heinen (1977) brought STS into the IS field in the very first volume of MIS Quarterly. Mumford developed ETHICS, a participatory design methodology, on the same foundation. The evidence is not ambiguous. The mechanisms are well understood. And yet every few years a new company rediscovers the longwall method, renames it "disruption" or "agility" or "moving fast," and acts surprised when the social system breaks.

I think the phrase itself gives the game away. "Move fast." What is being moved fast past? The parts of the work that involve other people. Deliberation, consent, negotiation, review, all the slow social processes that exist precisely because they prevent the kind of damage that fast technical optimization can cause. "Break things." The things being broken are always social things. Trust, relationships, institutions, norms, the shared agreements that make collective action possible. Nobody says "move fast and break the database." They say break things, because the things that break are the ones the optimization function does not measure.

Every time a tech company says we are moving fast, they mean we are not jointly optimizing the social subsystem. Trist and Bamforth could have predicted the outcome in 1951. The technologies change. The coal mine becomes a social network, a ride-hailing app, a blood-testing startup, a microblogging platform. The mechanism stays the same. An organization optimizes one subsystem in isolation, discovers the other subsystem was load-bearing, and then acts surprised when the roof falls in. What I want to see is a company that says we are moving at exactly the speed that allows both subsystems to evolve together. I am not holding my breath.