The Address Label Problem: Choosing a Good Document ID

Opening Scene

A moving company choosing how to label its storage locations faces a decision that seems minor at first but quietly shapes everything downstream: sequential numbers assigned in the order boxes arrive concentrate all of today’s activity in one small, congested part of the warehouse, while a well-distributed labeling scheme spreads activity evenly, avoiding that congestion entirely. The label itself doesn’t move any boxes. But the choice of labeling scheme genuinely determines how well the whole warehouse operates under real, sustained load.

Choosing a document ID carries this exact same outsized, easy-to-overlook impact.

In Plain English

A document’s ID — its primary identifier — isn’t just a label; in many document databases, it directly determines how documents get physically distributed across storage, particularly in a sharded (Article 12) system. A poorly chosen ID scheme, like a simple auto-incrementing sequence, can create a concentrated “hot spot” where all new writes land on the same physical location, while a well-chosen ID scheme distributes writes evenly across the system.

The Old Way

Choosing a genuinely good document ID scheme has always required thinking beyond simple uniqueness to real distribution and access patterns:

• A sequential, auto-incrementing ID is simple but risks write concentration, since every new document lands adjacent to the last one, creating a “hot spot” that can become a genuine performance bottleneck under high write volume, particularly in distributed systems.
• A well-distributed ID scheme — often based on a hash, a UUID, or a value genuinely spread across the ID space — avoids this concentration, spreading write load more evenly across a system’s underlying storage.
• The right choice genuinely depends on real access patterns: an ID scheme optimized purely for write distribution might make certain useful read patterns, like “give me the most recently created documents,” genuinely harder to serve efficiently, another real tradeoff to weigh deliberately.

Getting this right has always meant recognizing that a document’s ID is a genuine design decision with real performance consequences, not an arbitrary label to be chosen without thought.

What’s Changing (and Why AI Is the Reason)

1. AI-assisted ID scheme recommendation can analyze a collection’s expected write volume and access patterns, recommending an ID scheme genuinely suited to avoiding hot spots while still supporting the read patterns that actually matter. Rather than a modeler defaulting to a simple auto-incrementing ID without considering distribution consequences, AI-assisted analysis of anticipated load and query needs can inform a genuinely better-suited choice from the start.
2. AI-assisted hot spot detection can monitor a live system for write concentration caused by a poorly chosen ID scheme, catching a real performance problem that might otherwise only surface as a vague, hard-to-diagnose slowdown under load. This connects the theoretical ID scheme concern directly to observable, real production behavior, closing the gap between design-time theory and operational reality.
3. This concern becomes more consequential as AI agents generate a larger, less predictable volume of writes to a document database, since a write pattern an agent generates might not match the assumptions an ID scheme was originally designed around, making robust, evenly-distributing ID schemes a more valuable default. A system designed for a predictable, human-driven write pattern may need to be revisited as agent-driven write volume and patterns grow.

The Metaphor, Fully Extended

Moving Company Element	Document ID Concept
Storage locations labeled sequentially in arrival order, concentrating today’s activity in one area	A sequential, auto-incrementing document ID, risking write concentration in a distributed system
A well-distributed labeling scheme spreading activity evenly across the whole warehouse	A well-distributed ID scheme, based on a hash or UUID, spreading write load evenly
A labeling scheme that makes finding “the most recent arrivals” genuinely harder to do quickly	The real tradeoff between write-distribution and certain useful read patterns
A warehouse operations analyst studying exactly which sections experience congestion under heavy load	AI-assisted hot spot detection monitoring a live system for write concentration
A logistics planner choosing a labeling scheme upfront based on projected volume and how boxes will actually be retrieved	AI-assisted ID scheme recommendation informed by anticipated load and access patterns

Quick Recap

• A document’s ID isn’t just a label — in many document databases, it directly influences how documents get physically distributed across storage, particularly in sharded systems.
• Sequential IDs are simple but risk write concentration, while well-distributed IDs avoid that risk at some cost to certain useful read patterns.
• AI-assisted ID scheme recommendation can inform a genuinely well-suited choice from anticipated load and access patterns, and AI-assisted hot spot detection can catch write concentration in live systems.
• This concern grows more consequential as AI agents generate a larger, less predictable share of total write volume, potentially straining assumptions an ID scheme was originally designed around.

Where This Fits in the Series

Article 10 covered why a copy in every room is often the right, deliberate choice. This article covered the outsized impact of a document’s ID. Article 12 looks at what happens when the move outgrows one truck — sharding and horizontal scaling.