A best practice is what a past project would have killed for. It solved a real, bleeding problem — and then, years later, it is still there, solving no problem at all, quietly becoming the thing that breaks the next project. This is not a bug in a particular practice. It is what best practices are: crystallized local solutions, exported beyond the landscape that designed them.
Chesterton's fence is the warning on the other side: don't tear down a fence until you understand why it was built. Every best practice is such a fence. But the deeper problem is the reverse of Chesterton's — not the fence whose reason we've forgotten, but the fence that has outlived its reason. We draw it again and again, carefully, reverently, while the field it originally protected has long since been plowed flat.
A best practice is not wisdom. It is a local optimum: someone, somewhere, faced a messy tradeoff and found the cheapest solution that worked well enough. That solution got named, repeated, and finally enshrined as "the way we do things." Its triumph is that it bundled a hard-won insight into a repeatable ritual.
Its tragedy is the same act. Every practice encodes a tradeoff, and the memory of what it traded away is the first thing to erode. You inherit the practice intact; you never inherit the tradeoff. So when the world changes, you keep paying the cost the practice always carried — now for no benefit, because the benefit depended on a context that no longer exists.
This isn't a failure of discipline. Three structural forces guarantee that yesterday's practice becomes tomorrow's anti-pattern.
First, context drift. Microservices were a correct answer to the monolith that a ten-person team could no longer deploy. They became an anti-pattern only when they were applied, dogmatically, to systems where the monolith was never the bottleneck — where the real costs (distributed transactions, debugging across service boundaries, operational sprawl) outweighed a scaling problem that didn't exist. The practice didn't change; the situation did. The practice was exported; the situation wasn't.
Second, the friction between copying and understanding. The cheapest way to look rigorous is to copy the practice of a rigorous team. Cargo cults are not a curiosity; they are the default mode of learning. We adopt the ritual — the lint rules, the ceremony, the tool — and mistake the appearance of excellence for its substance. The practice becomes identity, and questioning it becomes heresy.
Third, Goodhart's law. Once a best practice becomes the metric people are judged by, it stops being a means and becomes an end. "We always do X" collapses into "you must be seen doing X." The ritual survives long after the outcome it indexed is gone, because the ritual is what gets scored.
The single most useful habit is to ask, of any practice: what did this trade away, and is that price still worth paying? Every rule — "never merge without review," "always pin every dependency," "normalize everything" — was chosen over a real alternative, and the choice had a cost. A team that can name the cost of its own practices understands them; a team that only recites them is a museum.
When the tradeoff price stays visible, two good things happen. First, you can recognize when the context has changed, because you are watching the price, not the ritual. Second, you can override the rule deliberately and in public — writing down why this case doesn't fit — which is the exact behavior dogmatism forbids and good engineering requires.
The distinction that saves you is between principles and practices.
The skillful move is to hold the principle and re-derive the practice for the present, instead of holding the practice and remembering the principle only as justification. That is the difference between an engineer who knows a rule and an engineer who understands it.
None of this is a license to rip out every practice — Chesterton's fence still stands for a reason, and the fool who tears it down without understanding will rebuild it at a worse cost. The mature stance is to keep both halves in tension: enough respect to learn why a practice exists, and enough nerve to doubt that its reason still holds. Today's best practice is tomorrow's anti-pattern precisely because rules are immortal while the problems they solve are not. The discipline is not in obeying the rule forever, and not in rejecting it on a whim — but in noticing, with uncomfortable honesty, the moment the field changed and the fence became just a fence.