Inevitable Memory Doom

Inevitable Memory Doom

No one could stop memory creep in the internet era and the attention era. No one can in the AI era.

A pattern has repeated like clockwork, twice. An industry floods a channel with content, storage becomes cheap enough that deletion stops being a habit. That leads to junk accumulating beyond a manageable threshold. The firms that sold you the content come in to help you cure it by selling additional memory.

Email was the first channel to run this cycle, photos and video ran it next, and enterprise software followed. AI coding agents too will run into this in a few years. Agentic coding creates a lot of intermediate files and never deletes them.

Email is the clearest example because the shift was so dramatic. Before 2004, Hotmail offered 2MB of storage and Yahoo Mail 4MB. Users routinely deleted old mail just to keep their inbox usable. Gmail launched with 1GB for free, about 500 times Hotmail's limit. Combined with fast search and threaded conversations, this changed the default behavior. People stopped filing and deleting. Search became the organizing system. Today, Google's 15GB shared limit across Gmail, Drive, and Photos has become the constraint. Google now monetizes that overflow through paid storage and AI bundles. The lever has changed, but the mechanism is similar. Remove the need to curate first, then charge when accumulation becomes expensive.

Photos and video followed the same cycle. Smartphone cameras made capture nearly free, while phone storage expanded rapidly. The norm changed from "keep the good ones" to "shoot everything, sort later." In practice, later rarely came. Libraries grew into tens or hundreds of gigabytes of mostly untouched content. Paid storage through Google One and iCloud+ then became increasingly normal. Users did not suddenly decide storage was valuable. They simply stopped deleting long before the bill arrived.

Enterprise code may now be entering the same loop. Agentic coding tools generate scratch files, CSVs, abandoned branches, logs, and other artifacts at much higher volume. Each file seems trivial, so teams have little incentive to clean continuously. Over time, the footprint compounds. Infrastructure vendors also benefit from greater storage consumption and larger software bundles. The pattern remains familiar. One product makes accumulation effortless. Another subscription eventually monetizes the consequence.

There is a broader irony here. Cheaper compute creates more data, so compute's win is also memory's win. Cameras created photo libraries. LLMs create generated text, code, logs, and artifacts. Agentic systems will create even more machine-generated exhaust. The same companies selling compute often sell the storage required to hold it.

This creates an opening for a different product. Instead of selling more storage or helping users clean up years of clutter, it could manage the footprint continuously. A second brain living in a user's ecosystem. It would identify stale, redundant, or low-value content as it accumulates. The problem is not that people are unwilling to clean. By the time they try, the pile is already too large to evaluate. It could turn a painful annual cleanup into an invisible background process. The opportunity is not a better closet. It is preventing the closet from filling up in the first place.