The context war

We won the war of gigabytes.

We are beginning the war of tokens.

For twenty years, our instinct was to keep everything. Storage became cheaper. Accumulation seemed free.

With AI, the same instinct is to send everything: the whole repository, the full history, every note, every conversation.

More context looks like more intelligence.

They are not the same thing.

Context is a scarce resource

Every useless token costs time, money and attention.

It dilutes what matters.

The new mastery is not opening the largest window available. It is choosing what deserves to enter it.

One precise file instead of an entire repository.

One summarised decision instead of three years of logs.

One clear question instead of an archive without hierarchy.

Preparing beats accumulating

Good context is not a mass of data.

It is compression without losing what matters.

Tomorrow, models will change again. The lasting skill will be to reread, remove, connect and transmit just enough.

Why do we still send an entire library when a summary would be enough?