AI tool cuts context window size by 40 percent

Who: Posted by @EnoReyes, who appears to work at Factory AI — a company selling infrastructure for AI-assisted software development. The underlying product announcement is from the Factory AI team, introducing a new feature called the Deferred Context Engine in their coding agent, Droid.

What's new: Factory AI has shipped a feature called the Deferred Context Engine inside Droid, their AI software development agent. The headline claim is a 40% reduction in context size by loading tools more selectively rather than all at once.

How it works: Most AI coding agents pre-load a large set of tools into context at the start of every task. Deferred Context Engine instead loads tools on demand — only pulling in what the agent actually needs for the next step. The analogy is keeping most of your toolbox in the van and only carrying the specific wrench you need onto the job site, rather than dragging the whole chest every time.

The numbers: The 40% context reduction is the concrete figure offered. Reyes argues this matters at scale because token volumes in a fully automated software pipeline run roughly 10 times higher than in a team where humans are still doing most of the work. A 5% efficiency gain at that volume, he writes, already translates to millions of dollars saved annually for large engineering organizations.

Caveats: The post is a sales pitch, not a technical paper, so the 40% figure comes without a controlled benchmark or independent verification. It is also unclear whether "context size" here means total tokens per task, per session, or something else — a meaningful distinction when calculating real cost savings. The Deferred Context Engine is presented as a differentiator over teams building their own agent harnesses from scratch, but the post does not compare against any named competitor.