Traffic: Why We Drive the Way We Do (and What It Says About Us)
Source: https://www.penguinrandomhouse.com/books/39036/traffic-by-tom-vanderbilt/ ↗
Vanderbilt uses driving as an empirical lens for complex systems: how individual behaviour aggregates into traffic patterns, how feedback loops produce congestion, how well-intentioned interventions often make things worse.
The chapter on late-merging versus early-merging — the science behind why zipper merges outperform "polite" ones — is worth the book on its own.
For product direction the value is the thinking it trains: most products are coordination problems at scale, and most interventions produce unintended consequences you can only see after aggregated behaviour has stabilised.
Vanderbilt writes for a general audience with rigour and humour.
Read it as a companion to Meadows's Thinking in Systems — the same lessons, made tangible by metal boxes moving fast.
Central argument
Vanderbilt argues that traffic is not primarily an engineering problem but a behavioural one: the aggregate of individually rational decisions — lane choices, following distances, merge timing — produces system-level outcomes that are often irrational, counterintuitive, and resistant to obvious fixes. His central finding is that congestion emerges from the interaction of small behaviours at scale, and that interventions designed to improve flow frequently backfire because they change the incentive landscape in ways designers didn't anticipate. The zipper merge is his sharpest illustration: the 'polite' early-merge norm that feels cooperative is empirically worse for throughput than the late merge that feels selfish.
Critique
Because Vanderbilt writes for a general audience, he tends to treat traffic research findings as more settled than the literature warrants — effect sizes, replication status, and the limits of specific study contexts are rarely interrogated. More substantively, the book was published before the large-scale arrival of GPS navigation and algorithmic routing, which have materially altered the feedback loops he describes: Waze-induced rat-running and dynamic rerouting create new emergent behaviours his framework gestures toward but cannot fully account for. A reader looking for a rigorous causal model of system intervention will find compelling analogies but not the formal machinery to act on them.
Why it matters for product
The core lesson — that most products are coordination problems whose emergent behaviour only becomes visible after aggregate usage stabilises — is a direct challenge to the product habit of evaluating features against intended use cases rather than against second-order interaction effects across the user population. Specifically, it reframes A/B testing: a metric lift in a controlled experiment can mask the congestion-equivalent that appears once the variant rolls out at scale and users start adapting to each other's adapted behaviour. For product directors, this is an argument for investing in system-level instrumentation and holding off on 'fixing' metrics that look broken but may be stabilising toward a new equilibrium.